Добірка наукової літератури з теми "Soft Biometry"

Objective: The objective of the study was to find whether incorporation of MTSTT in fetal weight estimation formulae which are traditionally based on biparietal diameter (BPD), head circumference (HC), abdominal circumference (AC), and femur length (FL) improves birth weight (BW) estimation. Methods: In a prospective observational study, MTSTT was measured within 1 week of delivery in 100 women with term singleton pregnancy along with other standard biometric parameters, i.e. BPD, HC, AC and FL, and MTSTT. Multiple regression analysis was carried out using PHOEBE regression software using different combinations of biometric variables to find out the best fit model of fetal weight estimation. The predicted BW was compared with actual neonatal BW soon after delivery and regression coefficients (R2) were determined for each of prediction models for comparing the accuracies. Results: Mean gestational age at delivery was 38.4±1.08 weeks and the BW of neonates varied between 2.18 kg and 4.38 kg (mean ± standard deviation: 3.07±0.43 kg). By adding MTSTT to BPD, HC, AC, and FL, we obtained the formula Log 10 (BW) = −0.14783+0.00725 *BPD +0.00043 *HC +0.00436 *AC +0.01942 *FL +0.16299 *MTSTT, which had a very good Pearson regression coefficient ((r2: 0.89 p<0.001) compared to conventional models based on standard fetal biometry. All prediction models had better strength of correlation when combined with MTSTT (p<0.001). The routine four parameter formula could identify 45% and 80% of fetuses within 5% and 10% weight range; pick up rate was further increased to 61% and 95% by addition of MTSTT. Conclusion: It is evident that addition of MTSTT to other biometric variables in models of fetal weight estimation improves neonatal BW prediction (r2=0.89).

<b><i>Objectives:</i></b> Fractional thigh volume (TVol) and fractional arm volume (AVol) measurements by three-dimensional (3D) ultrasound can reveal valuable information on fetal soft tissue development. However, it is not clear whether TVol or AVol provides better estimates of fetal body proportion and adiposity, independent of routine two-dimensional (2D) ultrasound biometry. The primary objective of the current study was to determine the correlations between fractional limb volumes (FLVs) and neonatal anthropometric parameters. <b><i>Design:</i></b> In this cross-sectional study, fetal FLVs were obtained within 24 h before term delivery from 40 medically and obstetrically uncomplicated pregnancies scheduled for elective cesarean section. TVol and AVol were determined using offline software. Postnatal morphometric data including birth weight; crown-heel, arm, and leg lengths; head, abdominal, mid-thigh, and mid-arm circumferences; and anterior thigh, biceps, and subscapular skinfold thicknesses were obtained. Pearson and partial correlation analyses were used to determine the relationships across antenatal volume calculations and neonatal indices. Correlation coefficients (<i>r</i>) were calculated. <b><i>Results:</i></b> Mean maternal age, BMI, and parity were 29.1 ± 5.4 years, 29.7 ± 3.5 kg/m², and 1.0 ± 1.3, respectively. AVol showed moderate correlations with most of the neonatal parameters, including mid-thigh circumference (<i>r</i> = 0.683), mid-arm circumference (<i>r</i> = 0.627), birth weight (<i>r</i> = 0.583), head circumference (HC, <i>r</i> = 0.560), and abdominal circumference (<i>r</i> = 0.542). However, TVol was weakly related to only some of the indices. After controlling for gestational age, maternal age, BMI, parity, and 2D ultrasound biometry, TVol was no longer associated with any of the parameters, while AVol was independently correlated with mid-thigh (<i>r</i> = 0.724) and mid-arm circumference (<i>r</i> = 0.560), birth weight (<i>r</i> = 0.502), ponderal index (<i>r</i> = 0.402), HC (<i>r</i> = 0.382), biceps (<i>r</i> = 0.384), and subscapular skinfold thickness (<i>r</i> = 0.350). <b><i>Limitations:</i></b> The current design includes limited number of pregnancies with only scheduled cesarean deliveries. Neonatal percent body fat was not calculated, and air-displacement plethysmography was not used to assess neonatal body composition. The study population was Caucasian with a relatively high maternal BMI, which may limit extrapolation of the results to other settings. <b><i>Conclusions:</i></b> AVoL measurements by 3D ultrasound before delivery are significantly correlated with most of the neonatal morphometric indices, independent of maternal characteristics and 2D biometric parameters. AVol may have advantages over TVol for assessing limb soft tissue development in term fetuses. Future research can focus on feasibility and predictive ability of AVol measurements in prospective studies that include serial biometry over time.

<b><i>Introduction:</i></b> We aim to evaluate the supplementary predictive value of soft tissue markers, including fetal limb volumes, for fetal birth weight and fat tissue weight. <b><i>Methods:</i></b> This is a prospective study of 60 patients undergoing term induction of labor. Ultrasound was performed 48 h before birth, and 2D sonographic measurements, subcutaneous tissue thickness, and 3D fetal limb volumes were taken. Birth weight and neonatal fat weight were assessed by plethysmography. Clinical data were collected. The relation between ultrasound and neonatal outcomes was assessed by univariate and multivariate predictive models. The estimated and actual birth weights were compared applying different published formulas, and systematic and random error were collected and compared. <b><i>Results:</i></b> 3D fetal limb volumes showed a strong relation to birth weight, absolute weight, and relative fat weight. The Lee 6 formula performed better than either Hadlock 3 or Lee 3 with the lowest random error. Fractional limb volumes involve a highly reproducible technique, with excellent correlation (intra-class coefficient &#x3e;0.90) for both inter- and intra-observer reliability. The prevalence of estimated EFW measures within 10% error from the actual birth weight was 71.7% with the Hadlock 3 model and 95.0% with the Lee 6 model (<i>p</i> = 0.09). <b><i>Conclusion:</i></b> Late assessment of 3D fetal limb volume in upper and lower extremities is not only useful for accurately predicting birth weight but is a useful marker for prediction of birth fat tissue weight.

Purpose: to evaluate dynamics of subjective and objective accommodation in children wearing bifocal soft contact lenses (BSCLs) for myopia control with +4.00 D addition power. M e t h o d s: the study involved 22 patients (44 eyes).Mean age amounted to 10.1 ± 1.46 years and mean myopic refraction amounted to –3.21 ± 1.23 D. Patients were fitted Prima BIO Bi-focal BSCLs (OKVision Retail, Russia). All patients underwent cycloplegic refraction assessed with Auto Ref/Keratometer ARK 530A (Nidek, Japan), had axial length measured with IOL Master 500 optical biometry device (Carl Zeiss, Germany), had positive relative accommodation (PRA) assessed with and without lenses and had binocular (BAR) and monocular (MAR) accommodative response assessed at a distance of 33 cm with WAM-5500 Binocular Accommodation Auto Ref/Keratometer (Grand Seiko, Japan) prior to wearing BSCLs as well as 3, 6 and 12 months after wearing BSCLs.Results: neither MAR nor BAR measured without lenses changed after 3, 6 and 12 months of wearing BSCLs (p >0,05). A change in PRA evaluated without lenses was noted after 12 months (p < 0,05). PRA evaluated with lenses after 3, 6 and 12 months differed from baseline significantly (p < 0,001). Over 12 months of wearing BSCLs, changes in AL (0.09 ± 0.17 mm) and cycloplegic refraction (0.3 ±0.43 D) correlated with baseline BAR and MAR loosely. C o n c l u s i o n: objective accommodation (MAR and BAR) did not change in the course of wearing BSCLs with +4.00 D addition power. Increase in PRA evaluated without BSCLs may be associated with improvement of accommodation due to a full correction in the optic zone. The gradual increase in PRA evaluated with BSCLs probably indicates an adaptation of patients to addition zone in near vision conditions.

A diagnosis was made on the fishing, processing, commercialization and generation of residues of the sururu fishery in Mundaú lagoon, Maceió-AL. Thirty nine fishermen were interviewed and also incursions were made for observations. Furthermore, sururus were collected to perform biometry. Residue of the sururu processing was collected to the fabrication of sururu waste meal (SWM), posteriorilly its bromatological composition was analysed. Fishermens of sururus fish 4.56 times week-1 and each fishing excursion lasts 5h26min, yielding 18.61 cans (18.61 to 37.22 kg soft tissue). Besides the sururus, 77.92% also catch fish, shrimp, and other organisms to complement the income. Average selling price was R$ 6.92 per kg of edible and CPUE from 3.42 to 6.85kg edible day-1. The average size and weight are 33.21 mm and 1.62 g, with 782.53 g of waste being generated to produce 1.0 kg of meat. SWM contained 37.00% calcium. It was verified that there is a need to create a management plan capable of reducing environmental impact in the lagoon, implement fishing programs and waste management, as well as improve the structure for the processing of that mollusc. Studies of the use of SWM as a source of calcium in diets are also recommended.

The present paper deals with the brief introduction of Turtle Rescue and Conservation Centre (TRCC) and description of some rescued turtles. Nine specimens of turtles belonging to five genera were rescued and translocated to the centre till date including 9.3 kg male Indian Peacock soft-shelled turtle (Nilssonia hurum) for the first time from Jhapa district. The rescue operations were conducted for the translocation of turtles confiscated from the local market, censorial collectors, fisherman and public residence. The high resolution photographs of captured specimens, their necessary biometry and GPS coordinates of location were taken. Species identification was done with the help of and pictorial field guide and relevant literatures. Climatic data of study area were recorded from Gainde Irrigation Project, Maidhar, Jhapa. Interviews were taken during field visits with the help of structured questionnaire. Preliminary rescue data showed that the Indian flap-shelled turtle (Lissemys punctata) and Yellow bellied roofed turtle (Pangshura flaviventer) were the most overexploited species in the vicinities of the study area. The climatic condition of the rescue centre and water quality found suitable to support terrestrial and freshwater turtles and other various wetland flora and fauna. However, the rapid population growth and habitat destruction due to deforestation, unmanaged urbanization and expansion of agricultural land are found as the major threats to the survival of turtles and other wetland creatures at the study area and its vicinities. DOI: http://dx.doi.org/10.3126/bibechana.v11i0.10394 BIBECHANA 11(1) (2014) 141-148

Background. Myopia is the most common clinical refractive error of the eye. Only in Russia, there are about 15 million myopic people. Currently, in addition to traditional correction (glasses, soft contact lenses), keratorefractive surgery actively develops; however, due to the initial parameters of the cornea or the magnitude of the refractive error, it may be contraindicated to the patient. Nowadays an alternative to these correction methods for young patients with refractive errors of high and ultra-high degrees is the implantation of a phakic intraocular lens.The aim. To evaluate the efficiency of refractive errors correction using phakic IOL implantation.Materials and methods. We carried out a retrospective analysis of outpatient records of 56 patients who received surgical treatment for myopia and complex myopic astigmatism at the Orenburg branch of the S. Fyodorov Eye Microsurgery Federal State Institution in the period from 2019 to 2020 (110 eyes), all patients underwent implantation of a phakic intraocular lens IPCL V2.0, IPCL V2.0 TORIC. Patients were examined before surgery and on the first day after surgery. In addition to standard, special clinical and functional examination methods were used: optical biometry on the IOL-Master 700, determination of the density of endothelial cells using an endothelial microscope, examination of the fundus under cycloplegia; if necessary, peripheral laser coagulation of the retina was performed before calculating phakic IOLs.Results. All operations and the early postoperative period went without complications. The target BCVA (0.75–1.0) was achieved in 90 (81.8 %) eyes. All patients were satisfied with the treatment results.Conclusion. Implantation of IPCL V2.0 and IPCL V2.0 TORIC phakic IOLs is an effective method for correcting refractive errors, regardless of the degree of myopia and the presence of astigmatism.

Biometric recognition is an advanced technology that employs physical features (such as fingerprint, iris and face capture) and behavioural features (such as gait, signature and voice) to identify people. Biometric features are reliable and valid ways to describe the unique properties of individuals, but there are often rigorous requirements on the position and characteristics of devices used for data acquisition. Since biometric features can be difficult to capture at a distance, soft biometric features, such as height, weight, skin colour and gender, have received much attention. Although the uniqueness of soft biometric features is not as intuitively obvious as traditional biometric features, numerous experiments have demonstrated that the desired recognition accuracy can be achieved by using different soft biometric features. This thesis will propose state-of-the-art multimodal biometric fusion techniques to improve recognition performance of soft biometrics. The first contribution of this thesis is to estimate fusion performance based on three types of soft biometrics - face, body and clothing. Feature level and score level fusion strategies will be employed to measure and analyse the influence of fusion on soft biometric recognition. The second key contribution of this research is that the analysis of the influence of distance on soft biometric traits and an exploration of the potency of recognition using fusion at varying distances have been performed. A new soft biometric database, containing images of the human face, body and clothing taken at three different distances, was created and used to obtain face, body and clothing attributes. First, this new database was constructed to explore the suitability of each modality at a distance: intuitively, the face is suitable for near field identification, and the body becomes optimal when the subject is further away. The new dataset is used to explore the potential of face, body and clothing for human recognition using fusion. In this section, some novel fusion techniques on different levels (feature, score and rank level) are proposed to improve soft biometric recognition performance. A Supervised Generalised Canonical Correlation (SG-CCA) methodology is proposed to fuse the soft biometric features. The proposed SG-CCA is numerically validated to be the best fusion method compared with other multi-modal fusion methods. An SVM-weighted Likelihood Ratio Test (SVM-LRT) method is proposed for score level fusion. The experimental results demonstrate that SVM-LRT-based fusion significantly outperforms the single-mode recognition. A novel joint density distribution-based rank-score fusion is also proposed to combine rank and score information. Analysis using the new soft biometric database demonstrates that recognition performance is significantly improved by using the new methods over single modalities at different distances.

La presente tesis se centra en la problemática existente a la hora de implementar un sistema de detección o clasificación binaria cuando es necesario combinar, integrar o fusionar diversas fuentes de información que pueden ser dependientes y heterogéneas entre sí. Las técnicas de fusión de datos tratan de combinar múltiples fuentes de información para alcanzar la exactitud y precisión en la toma de decisiones que no sería posible conseguir con el uso de una sola fuente de información de forma aislada. En un sistema de detección se pueden encontrar diferentes etapas y niveles de fusión: en la etapa de pre-detección encontramos los niveles de fusión de sensores y de características, donde se combinan los diferentes flujos de muestras proporcionados por una serie de sensores o diferentes características obtenidas del procesado estos; en la etapa de post- detección, se realiza la combinación de diferentes detectores, a través de la fusión de valoraciones continuas o de decisiones individuales aportadas por cada uno de ellos. Atendiendo al tipo de datos a combinar encontramos dos grupos: fusión soft, donde se combinan datos modelados mediante variables aleatorias continuas, caracterizadas mediante sus funciones de densidad de probabilidad (PDFs), o fusión hard, asociada a la combinación de las decisiones individuales tomadas en la etapa de fusión de detectores, donde se combinan datos binarios modelados mediante variables aleatorias discretas, caracterizadas por funciones de masa de probabilidad. Se destaca la fusión de scores como un caso particular de fusión soft asociada a la fusión de diversos detectores, en donde los datos a combinar presentan buenas propiedades discriminatorias de forma aislada y se encuentran definidos en un mismo rango normalizado [0,1]. En el presente trabajo se ha realizado una completa revisión del estado del arte en cuanto a técnicas de fusión y combinación de datos aplicadas en problemas de detección donde los datos pueden ser heterogéneos y dependientes entre sí. Se realiza una revisión en mayor profundidad de la técnica de estimación de PDFs basada en la teoría de cópulas, la cual puede ser usada en la fusión óptima de datos soft. Se destaca de forma especial tanto por su novedad e incipiente uso en el campo del procesado de señal, como por su adecuación en problemas de detección, permitiéndonos modelar de forma aislada las funciones marginales de los datos y la estructura de dependencia presente entre ellos, simplificando el problema de modelado de PDFs de datos heterogéneos y dependientes. Se ha propuesto una nueva técnica de fusión soft denominada integración-a, basada en una función de media-a, la cual, sin elevar mucho la complejidad, aporta un mayor grado de flexibilidad y de adaptación, siendo capaz de mejorar las prestaciones que se pueden obtener con respecto al resto de técnicas subóptimas utilizadas comúnmente en problemas de fusión de scores heterogéneos y dependientes entre sí. Se ha derivado un novedoso método de entrenamiento basado en el criterio de maximización parcial del área bajo la curva ROC. Se han utilizado diversas bases de datos públicas para poder testear y comprobar el correcto funcionamiento de las técnicas de fusión propuestas en problemas de autentificación multibiométrica. También se han aplicado algunas de las técnicas de fusión en la mejora de un sistema de detección de eventos acústicos. Se ha propuesto un nuevo tipo de detector basado en la teoría de cópulas denominado COCD para lidiar con el problema de la detección de señal desconocida en presencia de ruido aleatorio dependiente y no Gaussiano, centrándonos en su utilización para una aplicación de detección de eventos sonoros desconocidos. También se realiza un estudio de fusión de más de un canal de audio (utilizando más de un micrófono para captar diferentes señales) como método para incrementar las prestaciones obtenidas.
Soriano Tolosa, A. (2013). Fusión de datos estadísticamente dependientes en sistemas de detección [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/34780
TESIS

The human face carries some of the most individualizing features suitable for the personal identification. Facial morphology is used for the face matching of living. An extensive research is conducted to develop the matching algorithm to mimic the human ability to recognize and match faces. Human ability to recognize and match faces, however, is not errorless and it serves as the main argument precluding the visual facial matching from its use as an identification tool. The human face keeps its individuality after death. Compared to the faces of living, the faces of deceased are rarely used or researched for the face matching. Different factors influence the appearance of the face of the deceased compared to the face of the living, namely the early postmortem changes and decomposition process. On the other hand, the literature review showed the use of visual recognition in multiple cases of identity assessment after the natural disasters. Presented dissertation thesis is composed of several projects focused on the possibility of personal identification of the decedents solely based on the morphology of their face. Dissertation explains the need for such identification and explores the error rates of the visual recognition of deceased, the progress of facial changes due to the early decomposition and the possibility of utilization of soft biometric traits, specifically facial moles. Lastly, the dissertation presents the use of shape index (s) as a quality indicator of three different 3D scanners aimed towards the most suitable method for obtaining facial postmortem 3D images.

Ogni giorno, siamo esposti a varie immagini e video grazie ai social media, come Facebook, Youtube, Flickr, Instagram e altri. In questo scenario, l'esprimere preferenze per un dato contenuto multimediale (per esempio con l'uso del meccanismo di ``like'') è diventato pervasivo e imponente, diventando un fenomeno di massa sociale. Uno dei principali risultati nelle scienze cognitive è che i processi automatici di cui non siamo a conoscenza, modellano, per la maggior parte, la nostra percezione dell'ambiente. Il fenomeno si applica non solo al mondo reale, ma anche ai dati multimediali che consumiamo giornalmente. Ogni volta che osserviamo una immagine, guardiamo un video o ascoltiamo una registrazione, la nostra attenzione cosciente si concentra sul contenuto osservabile, ma la nostra cognizione percepisce spontaneamente intenzioni, opinioni, valori, attitudini e altri costrutti che, sebbene siano al di fuori della nostra consapevolezza cosciente, modellano le nostre reazioni e comportamenti. Finora, le tecnologie multimediali hanno trascurato questo fenomeno.Questa tesi discute il fatto che è possibile prendere in considerazione effetti cognitivi per migliore gli approcci multimediali. A questo scopo sono considerati principi di Computational Aesthetics e Social Signal Processing sotto un punto di vista computazionale. Da un lato la Computational Aesthetics ha la funzione di rendere applicabili decisioni estetiche in modo simile a come gli esseri umani sanno fare, permettendo alle tecnologie multimediali di modellare e valutare un senso comune della bellezza. Dall'altro lato il campo del Social Signal Processing ha lo scopo di modellare con algoritmi i processi cognitivi che codificano segnali sociali e che ci portano ad interagire in modo particolare con le persone o preferire immagini e video. Questa rappresenta una grande opportunità per la CA perché la risposta estetica umana è formata dalla combinazione di predisposizioni genetiche, assimilazione culturale e esperienze uniche individuali e in questo modo può essere imparata da immagini online usando la saggezza della folla.La tesi si focalizza sulle immagini come primo tentativo in questa direzione. Le motivazioni del perché concentrarsi sulle immagini sono molte: da un lato, scattare foto è una delle azioni comunemente svolte tramite l'uso di telefoni cellulari, e dell'altro lato, gli utenti postano online immagini originali o video e condividono e redistribuiscono quelli postati da altri.A questo scopo, la tesi presenta uno studio sull'estetica personale, dove lo scopo è quello di riconoscere le persone e le loro caratteristiche considerando le immagini che piacciono a queste sviluppando diversi approcci ibridi usando modelli generativi e di regressione. L'idea generale assume che, dato un insieme di immagini preferite, è possibile estratte un insieme di attributi che discriminano \textit{pattern} visuali, che possono essere usati per inferire caratteristiche personali del soggetto che le preferisce.Come primo contributo proponiamo un sistema di soft biometrics, che permette di discriminare un individuo rispetto ad altri usando le immagini che le/gli piacciono. Lo studio e sviluppo del sistema biometrico è diventato di primaria importanza sia per l'identificazione di individui che applicazioni di sicurezza è recommendation system. Su un dataset di 200 utenti e 40000 immagini, il sistema sviluppato raggiunge il 97\% di probabilità di indovinare l'utente corretto usando 5 immagini preferite come modello biometrico; per la capacità di verifica, l'EER è 0.11.Inoltre, abbiamo sviluppato un sistema capace di inferire la personalità di un soggetto usando le sue immagini preferite. La motivazione è che quando conosciamo una persona per la prima volta, tendiamo ad attribuire tratti di personalità ad esso/essa. Il processo è spontaneo e inconscio. Sebbene non necessariamente accurato, il processo comunque influenza significativamente il nostro comportamento nei confronti degli altri, specialmente quando si tratta di interazioni sociali. Il fenomeno è così diffuso che ha luogo non solo quando conosciamo altri in persona, ma anche quando li osserviamo in registrazioni video, o interagiamo con agenti artificiali che mostrano comportamenti simili agli umani o con materiale multimediale che le persone condividono online. Come risultato, la tesi mostra che ci sono pattern visuali che correlano con i tratti di personalità di utenti Flickr in misura statisticamente significativa, e che i tratti di personalità (sia auto valutati che attribuiti da altri) di questi utenti posso essere inferiti dalle immagini che questi ultimi marcano come preferite. Una della parti più importanti della tesi è stata la collezione del dataset PyschoFlickr, composto da 60000 immagini di 300 utenti Flickr annotate in termini di tratti di personalità sia auto attributi che attributi da 22 giudici. La predizione è eseguita usando più approcci (multiple instance regression e deep learning), raggiungendo una correlazione fino a 0.68 e un'accuratezza fino a 0.69 tra tratti reali e predetti.La predizione dei tratti attribuiti da altri ottiene risultati più alti rispetto a quelli auto attribuiti: la ragione è che le immagini dominano l'impressione della personalità che i giudici percepiscono e il consenso tra loro è statisticamente significativo. Questi due condizioni aiutano la regressione ad ottenere risultati più alti. Quando gli utenti auto giudicano la loro personalità, considerano anche altri informazioni che non sono disponibili nelle immagini che preferiscono, ad esempio, storia personale, la stato interiore, educazione, ecc.. Tuttavia, questo non permette di ottenere alti risultati nella regressione. Questo è un risultato importante che può aiutare a capire meglio il comportamento sociale delle persone a nel progettare agenti artificiali capaci di suscitare la percezione di tratti predefiniti desiderabili e fornire suggerimenti su come gestire le impressioni online usando le immagini preferite.
Everyday, we are exposed to various images and videos thanks to the social media, like Facebook, Youtube, Flickr, Instagram and others.In this scenario, the use of expressing preferences for a given multimedia content (for example by the use of liking mechanisms) has become pervasive and massive, becoming a social mass phenomenon.One of the main findings of cognitive sciences is that automatic processes of which we are unaware shape, to a significant extent, our perception of the environment. The phenomenon applies not only to the real world, but also to multimedia data we consume every day. Whenever we look at pictures, watch a video or listen to audio recordings, our conscious attention efforts focus on the observable content, but our cognition spontaneously perceives intentions, beliefs, values, attitudes and other constructs that, while being outside of our conscious awareness, still shape our reactions and behavior. So far, multimedia technologies have neglected such a phenomenon to a large extent. This thesis argues that taking into account cognitive effects is possible and it can also improve multimedia approaches. For this purpose we take into account Computational Aesthetics and Social Signal Processing principles under a computational point of view. On one side Computational Aesthetics makes applicable aesthetic decision in a similar fashion as human can allowing to multimedia technologies to learn, model and evaluate a common sense of beauty. On the other side,Social Signal Processing field has the aim of modeling with algorithms cognitive processes that codify social signal and that lead us to interact with a particular way with people or to prefer a particular image or video. This represents an invaluable opportunity for CA because human aesthetic response is formed by a combination of genetic predisposition, cultural assimilation, and unique individual experience and indeed it can be learned from online pictures using the wisdom of crowds.The thesis focuses on images as a first attempt in this direction.The motivation of why focusing on pictures are many: from one side, taking pictures is the action most commonly performed with mobile phones, on the other side, users either post online original images or videos or share and redistribute those posted by others. To this aim the thesis presents a study on personal aesthetics, where the goal is to recognize people and their characteristics by considering the images they like by developing several hybrid approaches using generative models and regressors.The general idea assumes that, given a set of preferred images, it is possible to extract a set of features individuating discriminative visual patterns, that can be used to infer personal characteristics of the subject that preferred them.As first contribution we propose a soft biometric system, that allows to discriminate an individual from another using the images he/she likes. The study and development of biometric system have become of paramount importance for both identification of individual and security applications and recommendation systems. On a dataset of 200 users and 40K images, the developed frameworks gives 97\% of probability of guessing the correct user using 5 preferred images as biometric template; as for the verification capability, the equal error rate is 0.11.Furthermore, we developed a system able to infer the personality of a subject using the images preferred by him/her. The motivation is that whenever we meet a person for the first time, but also when we observe her in video recordings, or we interact with an artifact displaying human-like behavior or with the multimedia material she shares online, we tend to attribute personality traits to her. The process is spontaneous and unconscious. While not necessarily accurate, the process still influences significantly our behavior towards others, especially when in comes to social interactions. As a supporting proof-of-concept, the thesis shows that there are visual patterns correlated with the personality traits of Flickr users to a statistically significant extent, and that the personality traits (both self-assessed and attributed by others) of those users can be inferred from the images these latter mark as ``favorite''. One of the most important part of the thesis has been the collection of the PsychoFlickr corpus, composed of 60K images of 300 Flickr users annotated in terms of personality traits both self and attributed by 22 assessors. The prediction are performed using multiple approaches (multiple instance regression approach and a deep learning framework), reaching a correlation up to 0.68 and an accuracy up to 0.69 between actual and predicted traits.The prediction of traits attributed from others achieve higher results compared to the self-assessed ones: the reason is that pictures dominate the personality impressions that the judges develop and the consensus across the judges is statistically significant. These two conditions help the regression approaches to achieve higher performances. When the users self-assess their personality, they take into account information that is not available in the favorite pictures like, e.g., personal history, inner state, education,etc.. Therefore, this does not allow the regression approaches to achieve high performances. This is an important finding as it can help to better understand the social behavior of people, to design artificial agents capable of eliciting the perception of predefined desirable traits and providing suggestions on how to manage online impressions using favorite pictures.

Thesis (Ph. D.)--University of Notre Dame, 2009.
Thesis directed by Kevin W. Bowyer for the Department of Computer Science and Engineering. "November 2009." Includes bibliographical references (leaves 153-157).

Biometric verification refers to the automatic verification of a person’s identity based on their behavioural and biological characteristics. Among various biometric modalities, the face is one of the most widely used since it is easily acquirable in unconstrained environments and provides a strong uniqueness. In recent years, face recognition systems spread world-wide and are increasingly involved in critical decision-making processes such as finance, public security, and forensics. The growing effect of these systems on everybody’s daily life is driven by the strong enhancements in their recognition performance. The advances in extracting deeply-learned feature representations from face images enabled the high-performance of current face recognition systems. However, the success of these representations came at the cost of two major discriminatory concerns. These concerns are driven by soft-biometric attributes such as demographics, accessories, health conditions, or hairstyles. The first concern is about bias in face recognition. Current face recognition solutions are built on representation-learning strategies that optimize total recognition performance. These learning strategies often depend on the underlying distribution of soft-biometric attributes in the training data. Consequently, the behaviour of the learned face recognition solutions strongly varies depending on the individual’s soft-biometrics (e.g. based on the individual’s ethnicity). The second concern tackles the user’s privacy in such systems. Although face recognition systems are trained to recognize individuals based on face images, the deeply-learned representation of an individual contains more information than just the person’s identity. Privacy-sensitive information such as demographics, sexual orientation, or health status, is encoded in such representations. However, for many applications, the biometric data is expected to be used for recognition only and thus, raises major privacy issues. The unauthorized access of such individual’s privacy-sensitive information can lead to unfair or unequal treatment of this individual. Both issues are caused by the presence of soft-biometric attribute information in the face images. Previous research focused on investigating the influence of demographic attributes on both concerns. Consequently, the solutions from previous works focused on the mitigation of demographic-concerns only as well. Moreover, these approaches require computationally-heavy retraining of the deployed face recognition model and thus, are hardly-integrable into existing systems. Unlike previous works, this thesis proposes solutions to mitigating soft-biometric driven bias and privacy concerns in face recognition systems that are easily-integrable in existing systems and aim for more comprehensive mitigation, not limited to pre-defined demographic attributes. This aims at enhancing the reliability, trust, and dissemination of these systems. The first part of this work provides in-depth investigations on soft-biometric driven bias and privacy concerns in face recognition over a wide range of soft-biometric attributes. The findings of these investigations guided the development of the proposed solutions. The investigations showed that a high number of soft-biometric and privacy-sensitive attributes are encoded in face representations. Moreover, the presence of these soft-biometric attributes strongly influences the behaviour of face recognition systems. This demonstrates the strong need for more comprehensive privacy-enhancing and bias-mitigating technologies that are not limited to pre-defined (demographic) attributes. Guided by these findings, this work proposes solutions for mitigating bias in face recognition systems and solutions for the enhancement of soft-biometric privacy in these systems. The proposed bias-mitigating solutions operate on the comparison- and score-level of recognition system and thus, can be easily integrated. Incorporating the notation of individual fairness, that aims at treating similar individuals similarly, strongly mitigates bias of unknown origins and further improves the overall-recognition performance of the system. The proposed solutions for enhancing the soft-biometric privacy in face recognition systems either manipulate existing face representations directly or changes the representation type including the inference process for verification. The manipulation of existing face representations aims at directly suppressing the encoded privacy-risk information in an easily-integrable manner. Contrarily, the inference-level solutions indirectly suppress this privacy-risk information by changing the way of how this information is encoded. To summarise, this work investigates soft-biometric driven bias and privacy concerns in face recognition systems and proposed solutions to mitigate these. Unlike previous works, the proposed approaches are (a) highly effective in mitigating these concerns, (b) not limited to the mitigation of concerns origin from specific attributes, and (c) easily-integrable into existing systems. Moreover, the presented solutions are not limited to face biometrics and thus, aim at enhancing the reliability, trust, and dissemination of biometric systems in general.

Traditionally, recognition systems were only based on human hard biometrics. However, the ubiquitous CCTV cameras have raised the desire to analyze human biometrics from far distances, without people attendance in the acquisition process. Highresolution face closeshots are rarely available at far distances such that facebased systems cannot provide reliable results in surveillance applications. Human soft biometrics such as body and clothing attributes are believed to be more effective in analyzing human data collected by security cameras. This thesis contributes to the human soft biometric analysis in uncontrolled environments and mainly focuses on two tasks: Pedestrian Attribute Recognition (PAR) and person reidentification (reid). We first review the literature of both tasks and highlight the history of advancements, recent developments, and the existing benchmarks. PAR and person reid difficulties are due to significant distances between intraclass samples, which originate from variations in several factors such as body pose, illumination, background, occlusion, and data resolution. Recent stateoftheart approaches present endtoend models that can extract discriminative and comprehensive feature representations from people. The correlation between different regions of the body and dealing with limited learning data is also the objective of many recent works. Moreover, class imbalance and correlation between human attributes are specific challenges associated with the PAR problem. We collect a large surveillance dataset to train a novel gender recognition model suitable for uncontrolled environments. We propose a deep residual network that extracts several posewise patches from samples and obtains a comprehensive feature representation. In the next step, we develop a model for multiple attribute recognition at once. Considering the correlation between human semantic attributes and class imbalance, we respectively use a multitask model and a weighted loss function. We also propose a multiplication layer on top of the backbone features extraction layers to exclude the background features from the final representation of samples and draw the attention of the model to the foreground area. We address the problem of person reid by implicitly defining the receptive fields of deep learning classification frameworks. The receptive fields of deep learning models determine the most significant regions of the input data for providing correct decisions. Therefore, we synthesize a set of learning data in which the destructive regions (e.g., background) in each pair of instances are interchanged. A segmentation module determines destructive and useful regions in each sample, and the label of synthesized instances are inherited from the sample that shared the useful regions in the synthesized image. The synthesized learning data are then used in the learning phase and help the model rapidly learn that the identity and background regions are not correlated. Meanwhile, the proposed solution could be seen as a data augmentation approach that fully preserves the label information and is compatible with other data augmentation techniques. When reid methods are learned in scenarios where the target person appears with identical garments in the gallery, the visual appearance of clothes is given the most importance in the final feature representation. Clothbased representations are not reliable in the longterm reid settings as people may change their clothes. Therefore, developing solutions that ignore clothing cues and focus on identityrelevant features are in demand. We transform the original data such that the identityrelevant information of people (e.g., face and body shape) are removed, while the identityunrelated cues (i.e., color and texture of clothes) remain unchanged. A learned model on the synthesized dataset predicts the identityunrelated cues (shortterm features). Therefore, we train a second model coupled with the first model and learns the embeddings of the original data such that the similarity between the embeddings of the original and synthesized data is minimized. This way, the second model predicts based on the identityrelated (longterm) representation of people. To evaluate the performance of the proposed models, we use PAR and person reid datasets, namely BIODI, PETA, RAP, Market1501, MSMTV2, PRCC, LTCC, and MIT and compared our experimental results with stateoftheart methods in the field. In conclusion, the data collected from surveillance cameras have low resolution, such that the extraction of hard biometric features is not possible, and facebased approaches produce poor results. In contrast, soft biometrics are robust to variations in data quality. So, we propose approaches both for PAR and person reid to learn discriminative features from each instance and evaluate our proposed solutions on several publicly available benchmarks.
This thesis was prepared at the University of Beria Interior, IT Instituto de Telecomunicações, Soft Computing and Image Analysis Laboratory (SOCIA Lab), Covilhã Delegation, and was submitted to the University of Beira Interior for defense in a public examination session.

AbstractWith the proliferation of facial analytics and automatic recognition technology that can automatically extract a broad range of attributes from facial images, so-called soft-biometric privacy-enhancing techniques have seen increased interest from the computer vision community recently. Such techniques aim to suppress information on certain soft-biometric attributes (e.g., age, gender, ethnicity) in facial images and make unsolicited processing of the facial data infeasible. However, because the level of privacy protection ensured by these methods depends to a significant extent on the fact that privacy-enhanced images are processed in the same way as non-tampered images (and not treated differently), it is critical to understand whether privacy-enhancing manipulations can be detected automatically. To explore this issue, we design a novel approach for the detection of privacy-enhanced images in this chapter and study its performance with facial images processed by three recent privacy models. The proposed detection approach is based on a dedicated attribute recovery procedure that first tries to restore suppressed soft-biometric information and based on the result of the restoration procedure then infers whether a given probe image is privacy enhanced or not. It exploits the fact that a selected attribute classifier generates different attribute predictions when applied to the privacy-enhanced and attribute-recovered facial images. This prediction mismatch (PREM) is, therefore, used as a measure of privacy enhancement. In extensive experiments with three popular face datasets we show that the proposed PREM model is able to accurately detect privacy enhancement in facial images despite the fact that the technique requires no supervision, i.e., no examples of privacy-enhanced images are needed for training.

Soft and stretchable semiconductors could enable seamless integration of wearable or implantable sensors with biological systems. We report on flexible and stretchable organic photodiodes with silicon-photodetector like performance and applications in biometric monitoring.