Search

The harm caused by deepfake face images is increasing. To proactively defend against this threat, this paper innovatively proposes a destructive active defense algorithm for deepfake face images (DADFI). This algorithm adds slight perturbations to the original face images to generate adversarial samples. These perturbations are imperceptible to the human eye but cause significant distortions in the outputs of mainstream deepfake models. Firstly, the algorithm generates adversarial samples that maintain high visual fidelity and authenticity. Secondly, in a black-box scenario, the adversarial samples are used to attack deepfake models to enhance their offensive capabilities. Finally, destructive attack experiments were conducted on the mainstream face datasets CASIA-FaceV5 and CelebA. The results demonstrate that the proposed DADFI algorithm not only improves the generation speed of adversarial samples but also increases the success rate of active defense. This achievement can effectively reduce the harm caused by deepfake face images. [ABSTRACT FROM AUTHOR]

Biometric systems have recently become popular for modern security applications. Unfortunately, these systems have been the target of many hacking attempts. Biometrics in biometric databases will be lost forever if they are hacked and stolen. As a result, there is an urgent need to implement modern upgradable biometric systems. Cancelable biometrics is based on the principle of transforming biometric data to alternate templates that cannot be easily exploited by an impostor or attacker and can be discarded if violated. The idea in this paper depends on the inverse filter in a cancelable face recognition system. Masked biometric images are produced in the proposed cancelable face recognition system by blurring, noise addition, and then inverse filtering. In image restoration theory, it is well understood that inverse filtering contributes to noise enhancement, which is an undesirable effect. On the contrary, this result would be desired in cancelable biometric systems. When noise is magnified, it can obscure the original biometrics, resulting in cancelable templates. This is the philosophy that underpins the proposed system. The proposed system was tested on the Olivetti and Oracle (ORL) dataset, the Labeled Faces in the Wild (LFW) database, and the Face Recognition Technology (FERET) database. Simulation results using evaluation metrics such as non-invertibility, unlinkability, visual inspection, false positive rate, false negative rate, Equal Error Rate (EER), decidability, correlation coefficient, and Area under the Receiver Operating Characteristic curve (AROC) show that the proposed cancelable biometric recognition system is quite effective for several security applications. [ABSTRACT FROM AUTHOR]

Copyright of Przegląd Elektrotechniczny is the property of Przeglad Elektrotechniczny and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Semantic super-resolution (SR) is an approach that improves the SR performance by leveraging semantic information about the scene. This study develops a novel semantic SR method that is based on the generative adversarial network (GAN) framework and self-distillation. A discriminator is adversarially trained along with a generator to extract semantic features from images and distinguish semantic differences between images. To train the generator, an additional adversarial loss is computed from the discriminator’s outputs of SR images belonging to the same category and minimized via self-distillation. This guides the generator to learn implicit category-specific semantic priors. We conducted experiments for SR of text and face images using the Enhanced Deep Super-Resolution (EDSR) generator and the SRGAN discriminator. Experimental results showed that our method can contribute to improving both the quantitative and qualitative quality of SR images. Although the improvement varied depending on image category and dataset, the peak signal-to-noise ratio (PSNR) value increased by up to 0.87 dB and the perceptual index (PI) decreased by up to 0.17 by using our method. Our method outperformed existing semantic SR methods.

Face synthesis and editing has increased cumulative consideration by the improvement of generative adversarial networks (GANs). The proposed attentional GAN-deep attentional multimodal similarity modal (AttnGANDAMSM) model focus on generating high-resolution images by removing discriminator components and generating realistic images from textual description. The attention model creates the attention map on the image and automatically retrieves the features to produce various sub-areas of the image. The DAMSM delivers fine-grained image-text identical loss to generative networks. This study, first describe text phrases and the model will generate a photorealistic high-resolution image composed of features with high accuracy. Next, model will fine-tune the selected features of face images and it will be left to the control of the user. The result shows that the proposed AttnGAN-DAMSM model delivers the performance metrics like structural similarity index measure (SSIM), feature similarity index measure (FSIM) and frechet inception distance (FID) using CelebA and CUHK face sketch (CUFS) dataset. For CelebFaces attribute (CelebA) dataset, the SSIM achieves 78.82% and for CUFS dataset, the SSIM achieves 81.45% which ensures accurate face synthesis and editing compared with existing methods such as GAN, SuperstarGAN and identity-sensitive GAN (IsGAN) models. [ABSTRACT FROM AUTHOR]

Age estimation is a difficult process as it is impacted by race, gender, internal and external characteristics. The performance of age estimation was lower in order to accurately recognize the age of face images with minimal time. In order to overcome these issues, the Adaptive Deep Recurrent Brown Boosted Softmax Regressive Classification (ADRBSRC) method is proposed for aging facial recognition. Initially, the ADRBSRC method performs preprocessing using the adaptive bilateral filtering method, which is employed to eliminate the noise in the images. The ADRBSRC approach then extracts the important features of the input facial image using Adaptive Deep Recurrent Feature Learning (ADRFL) technique. Then, this method employs an ensemble learning method called Brown Boosted Softmax Regressive Classifier (BBSRC) in which each input image is classified into multiple age group classes (i.e. childhood age, teenage, young age, middle age, and old age) by designing a strong classifier. When compared to conventional methods like ensemble learning and ensemble CNN2ELM, the experimental results of the ADRBSRC method show that it improves Recognition Accuracy (RA) by 21% and 12% in the FGNET database, 24% and 13% in the MORPH database, 23% and 14% in the AGFW database, and 25% and 19% in the CALFW database. It reduces the computational time (CT) by 12% and 19% in the FGNET database, and 13% and 21% in the MORPH database, and in the AGFW database by 12% and 20%, and by 14% and 22% in the CALFW database. [ABSTRACT FROM AUTHOR]

Face image inpainting is a hot topic of image processing research in recent years.Due to the loss of excessive sematic information,it is a difficult problem to inpaint large area missing of face images.Aiming at the problem of inpainting face images,a step-by-step image inpainting algorithm based on generative adversarial network is proposed.Face images inpainting task is divided into two steps.Firstly,face images are completed through the pre-completion network,and pre-completion images is enhanced feature through the enhancement network.The discriminator judges the difference between the pre-completion images,the enhanced images and the ideal image respectively.The long-term memory unit is used to connect the information flow of two parts.Secondly,the adversarial loss,content loss and total variation loss are combined to improve the effectively.Experiments are conducted on CelebA dataset,and this algorithm has an improvement of 16.84%~22.85% in PSNR and 10%～12.82% in SSIM compared with others typical image inpainting algorithms

Face images captured through glass are usually contaminated by reflections. The low-transmitted reflections make the reflection removal more challenging than for general scenes because important facial features would be completely occluded. In this paper, we propose and solve the face image reflection removal problem. We recover the important facial structures by incorporating inpainting ideas into a guided reflection removal framework, which takes two images as the input and considers various face-specific priors. We use a newly collected face reflection image dataset to train our model and compare with state-of-the-art methods. The proposed method shows advantages in estimating reflection-free face images for improving face recognition. [ABSTRACT FROM AUTHOR]

Image synthesis using representations learned by deep neural networks has gained wide attention in recent years. Among the different categories of natural images, face images are very important because of their broad range of applications. However, it is very challenging to synthesize face images due to their highly complicated hierarchical structure and the uniqueness of information contained in individual face images. This paper aims at providing a comprehensive review of the recent developments and applications of face synthesis and semantic manipulations using deep learning and discusses future perspectives for improving face perception. [ABSTRACT FROM AUTHOR]

Mobile devices have their popularity and affordability greatly increased in recent years. As a consequence of their ubiquity, these devices now carry all sorts of personal data that should be accessed only by their owner. Even though knowledge‐based procedures are still the main methods to secure the owner's identity, recently biometric traits have been employed for more secure and effortless authentication. In this work, the authors propose a facial verification method optimised to the mobile environment. It consists of a two‐tiered procedure that combines hand‐crafted features and a convolutional neural network (CNN) to verify if the person depicted in a photograph corresponds to the device owner. To train a CNN for the verification task, the authors propose a hybrid‐image input, which allows the network to process encoded information of a pair of face images. The proposed experiments show that the solution outperforms state of the art face verification methods, providing a 4× speedup when processing an image in recent smartphone models. Additionally, the authors show that the two‐tiered procedure can be coupled with existing face verification CNNs improving their accuracy and efficiency. They also present a new data set of selfie pictures – RECOD Selfie data set – that hopefully will support future research in this scenario. [ABSTRACT FROM AUTHOR]

The main objective of this study is to compare face recognition accuracies in the case when the grey levels in each pixel of the face images are replaced by the gradient and the surface normal vectors. Extensive information is provided to explain the differences between the gradient and the proposed features. Some well‐known face recognition methods, such as common vector approach (CVA), discriminative CVA, and support vector machines are applied to the well‐known databases of AR and Yale for comparison other than introducing a new method what the authors called as Sum of Pixel Slope Similarities Approach. The authors' experimental results are compared with the state‐of‐the‐art methods to the best of their knowledge. In conclusion, their results imply that using the surface normal vectors rather than the gradient vectors in each pixel with no additional work on their elements gives better recognition rates. [ABSTRACT FROM AUTHOR]

Utilising different representations of face images is known to be helpful in face recognition. Inthis study,theauthorspropose two fusion techniques that make use of multiple face image features bycollaboratively training a similarity metric learner, based on Siamese neuralnetworks. This training procedure takes two (or possibly more) features of twoface images and outputs a similarity score that depicts whether the faces belongto the same person or not.Theauthorsinvestigate two approaches of collaborative similarity metric learning (CoSiM),both of which are based on training Siamese neural networks jointly, as a meansof early fusion. The experiments are employed on hand‐crafted features such asscale‐invariant feature transform (SIFT) and variants of the local binarypattern (LBP), on the YouTube Faces and the Labeled Faces in the Wild data sets.The authors provide theoretical and empirical comparisons of the proposed modelsagainst the related methods in the literature. It is shown that the proposedtechnique improves on the verification accuracy, compared to singlefeature‐based baselines. By only utilising simple features like SIFT and LBP,the proposed techniques are shown to yield comparable results to the state ofthe art techniques, which depend on deep convolutional architectures or higherlevel features. [ABSTRACT FROM AUTHOR]

The authors propose an adaptive face recognition algorithm based on the discrete cosine transform (DCT) coefficients approach. For the database's establishment, the face images are pre‐processed with colour transform, hair cutting, and background removing to eliminate non‐face information. The recognised kernel applied the weights of DCT coefficient distribution with the entire image transformation, to avoid position mismatch and reduce the light effect. The key coefficients of DCT are chosen from the training database by maximum variance. The fast search mode can reject 90% weak candidates with few coefficients to fasten the processing speed. The significant coefficients weighting methods are used to enhance face features. Only using 50 coefficients per picture, the recognition rate can achieve 95% for ORL face database testing. For real‐time recognition, camera imaging is processed with algorithms using C‐programming based on Windows system. The recognition rate can achieve 95% and the speed is about nine frames per second for real‐time recognition in practice. [ABSTRACT FROM AUTHOR]

For face recognition, some very large‐scale datasets are publicly available in recent years, which are usually collected from the Internet using search engines, and thus have many faces with wrong identity (ID) labels (outliers). Additionally, the face images in these datasets have different qualities because of uncontrolled situations. The authors propose a novel approach for cleaning the ID label error, handling face images in different qualities. The face ID labels cleaned by their method can train better models for low‐quality face recognition since more low‐quality images are correctly labelled for training a deep model. In their low‐to‐high‐quality face verification experiments, the deep model trained on their cleaning results of MS‐Celeb‐1M.v1 face dataset outperforms the same model trained on the same dataset cleaned by the semantic bootstrapping method. They also apply their ID label cleaning method on a subset of the cross‐age celebrity dataset (CACD) face dataset, in which their quality‐based cleaning can deliver higher precision and recall than a previous method on detecting the ID label errors. [ABSTRACT FROM AUTHOR]

The capacity to recognize perceptually similar complex visual stimuli such as human faces has classically been thought to require a large primate, and/or mammalian brain with neurobiological adaptations. However, recent work suggests that the relatively small brain of a paper wasp, Polistes fuscatus, possesses specialized face processing capabilities. In parallel, the honeybee, Apis mellifera, has been shown to be able to rely on configural learning for extensive visual learning, thus converging with primate visual processing. Therefore, the honeybee may be able to recognize human faces, and show sophisticated learning performance due to its foraging lifestyle involving visiting and memorizing many flowers. We investigated the visual capacities of the widespread invasive wasp Vespula vulgaris, which is unlikely to have any specialization for face processing. Freely flying individual wasps were trained in an appetitive-aversive differential conditioning procedure to discriminate between perceptually similar human face images from a standard face recognition test. The wasps could then recognize the target face from novel dissimilar or similar human faces, but showed a significant drop in performance when the stimuli were rotated by 180°, thus paralleling results acquired on a similar protocol with honeybees. This result confirms that a general visual system can likely solve complex recognition tasks, the first stage to evolve a visual expertise system to face recognition, even in the absence of neurobiological or behavioral specialization.

One of the challenges of mobile health is to provide a way of maintaining privacy in the access to the data. Especially, when using ICT for providing access to health services and information. In these scenarios, it is essential to determine and verify the identity of users to ensure the security of the network. A way of authenticating the identity of each patient, doctor or any stakeholder involved in the process is to use a software application that analyzes the face of them through the cams integrated in their devices. The selection of an appropriate facial authentication software application requires a fair comparison between alternatives through a common database of face images. Users usually carry out authentication with variations in their aspects while accessing to health services. This paper presents both 1) a database of facial images that combines the most common variations that can happen in the participants and 2) an algorithm that establishes different levels of access to the data based on data sensitivity levels and the accuracy of the authentication.

Subject of Research. The paper deals with methods and algorithms for mutual transformation of related pairs of images in order to enhance the capabilities of cross-modal multimedia retrieval (CMMR) technologies. We have thoroughly studied the problem of mutual transformation of face images of various kinds (e.g. photos and drawn pictures). This problem is widely represented in practice. Research is this area is based on existing datasets. The algorithms we have proposed in this paper can be applied to arbitrary pairs of related images due to the unified mathematical specification. Method. We have presented three image transformation algorithms. The first one is based on principal component analysis and Karhunen-Loève transform (1DPCA/1DKLT). Unlike the existing solution, it does not use the training set during the transformation process. The second algorithm assumes generation of an image population. The third algorithm performs the transformation based on two-dimensional principal component analysis and Karhunen-Loève transform (2DPCA/2DKLT). Main Results. The experiments on image transformation and population generation have revealed the main features of each algorithm. The first algorithm allows construction of an accurate and stable model of transition between two given sets of images. The second algorithm can be used to add new images to existing bases and the third algorithm is capable of performing the transformation outside the training dataset. Practical Relevance. Taking into account the qualities of the proposed algorithms, we have provided recommendations concerning their application. Possible scenarios include construction of a transition model for related pairs of images, mutual transformation of the images inside and outside the dataset as well as population generation in order to increase representativeness of existing datasets. Thus, the proposed algorithms can be used to improve reliability of face recognition performed on images of various kinds. Moreover, these techniques can be applied to address a wide variety of other CMMR problems.

Many pieces of research have been conducted on image‐restoration techniques to recover high‐quality images from their low‐quality versions, but they usually aim to handle a single degraded factor. However, captured images usually suffer from various degradation factors, such as low resolution and compression distortion, in the procedures of image acquisition, compression, and transmission simultaneously. Ignoring the correlation of different degraded factors may result in the limited efficiency of the existing image‐restoration methods for captured images. A joint deep‐network‐based image‐restoration algorithm is proposed to establish a restoration framework for image deblocking and super‐resolution. The proposed convolutional neural network is made up of two stages. A deblocking network is constructed with two cascade deblocking subnets first, then, super‐resolution is performed by a very deep network with skipping links. Cascading these two stages forms a novel deep network. An end‐to‐end training scheme is developed, which makes the two stages be trained jointly so as to achieve better performance. Intensive evaluations have been conducted to measure the performance of the authors' method both in general images and face images. Experimental results on several datasets demonstrate that the proposed method outperforms other state‐of‐the‐art methods, in terms of both subjective and objective performances. [ABSTRACT FROM AUTHOR]

Automatic facial kinship verification is a challenging topic in computer vision due to its complexity and its important role in many applications such as finding missing children and forensics. This paper presents a Facial Kinship Verification (FKV) approach based on an automatic and more efficient two-step learning into color/texture information. Most of the proposed methods in automatic kinship verification from face images consider the luminance information only (i.e. gray-scale) and exclude the chrominance information (i.e. color) that can be helpful, as an additional cue, for predicting relationships. We explore the joint use of color-texture information from the chrominance and the luminance channels by extracting complementary low-level features from different color spaces. More specifically, the features are extracted from each color channel of the face image and fused to achieve better discrimination. We investigate different descriptors on the existing face kinship databases, illustrating the usefulness of color information, compared with the gray-scale counterparts, in seven various color spaces. Especially, we generate from each color space three subspaces projection matrices and then score fusion methodology to fuse three distances belonging to each test pair face images. Experiments on three benchmark databases, namely the Cornell KinFace, the KinFaceW (I & II) and the TSKinFace database, show superior results compared to the state of the art. [ABSTRACT FROM AUTHOR]

Face pose estimation from 2D images is an important topic in the field of computer vision. However, the distribution of face images, under pose variations, is highly nonlinear and complex. We deal with this problem based on the following understanding: (1) the essence of face pose estimation is to assign a face image to a pose template, and the variance of pose templates should be as large as possible; (2) kernel trick is a good auxiliary tool for describing nonlinear distribution. In this paper, we propose a face pose estimation method based on maximum separability of pose templates and incorporate the kernel technique to help model the nonlinear distribution of face patterns under pose variations. The proposed method involves a kernel subspace projection phase and the nearest neighbor classification. Experimental results on the CMU PIE face database and the UMIST database show that the proposed method can achieve high pose estimation performance. [ABSTRACT FROM AUTHOR]

Face recognition from side‐view positions is an essential task for recognition systems with real‐world scenarios. Most of the existing face recognition methods rely on alignment of face images into some canonical form. However, alignment in side‐view faces can be challenging due to lack of symmetry and a small number of reliable reference points. To the best of the author's knowledge, only a few of the existing methods deal with video‐based face recognition from side‐view images, and not many databases include sufficient video footage to study this task. Here, the authors propose an automatic side‐view face recognition system designed for home safety applications. They first contribute a newly collected video face database, named UT‐DOOR, where 98 subjects were recorded with four cameras attached at doorposts as they pass through doors. Secondly, they propose a face recognition system, where they automatically detect and recognise faces using side‐view images in videos. One of the attractive properties of this system is that they use cameras with limited view angle to preserve the privacy of the people. They review several databases and test their system both on the CMU Multi‐PIE database and the UT‐DOOR database for comparison. Experimental results show that their system can successfully recognise side‐view faces from videos. [ABSTRACT FROM AUTHOR]

Featured Application: Facial images provide variety of information about a person such as personal identity, age, gender, color, ethnicity, head pose, eye gaze, and emotion. Age estimation can be used to build real-world applications in biometrics, law enforcement, surveillance, and human-computer interaction (HCI). Automatic age estimation from unconstrained facial images is a challenging task and it recently has gained much attention due to its wide range of applications. In this paper, we propose a new model based on convolutional neural networks (CNNs) and l2,1-norm to select age-related features for the age estimation task. A new cost function is proposed. To learn and train the new model, we provide the analysis and the proof for the convergence of the new cost function to solve minimization problem of deep neural networks (DNNs) and the l2,1-norm. High-level features are extracted from the facial images by using transfer learning, since there are currently not enough large age databases that can be used to train a deep learning network. Then, the extracted features are fed to the proposed model to select the most efficient age-related features. In addition, a new system that is based on DNN to jointly fine-tune two different DNNs with two different feature sets is developed. Experimental results show the effectiveness of the proposed methods and achieved the state-of-art performance on a public database. [ABSTRACT FROM AUTHOR]

Autoassociative neural networks can be used for nonlinear processing and normalization of data. Because, firstly, they are able to learn and simulate complex nonlinear communications, and secondly, the communications can be learned through analyzing and distributing information on the neurons and weights and then combining the results of their processing. In this way, they actually make an interpolation between the input data and their communications. But these neural networks cannot model attractor dynamics that is obviously used in brain function. In this paper, the output of autoassociative neural network is connected to its input, and through recursive connections the ability of attractor behavior in these models is provided. This study showed that a recursive neuron with a logistic function forms two attractors, in its training point and its symmetry, but for it with a sigmoid nonlinear function can be formed an attractor in a certain range. In the experiments on face images, it was shown that the absorbance of the images to their attractors was improved from 52.67% to 87.27% by increasing the number of layers and the supervised layer-by-layer pre-training in order to adjust the attractors. [ABSTRACT FROM AUTHOR]

The recent significant progress in face recognition is mainly achieved using learning‐based (LE) techniques via an exhaustive training involving a huge number of face samples. However, in many applications, the number of face images available for training may be very limited. This makes LE techniques impractical for learning discriminative features and models. Thus, limited number of face samples (i.e. scarce data) degrades the recognition performance of most existing methods. To overcome this problem, the authors propose a novel approach based on two‐layer collaborative representation to exploit the abundance of samples in some classes to enrich the scarce data in other classes. The first‐layer collaborative representation uses the abundance of samples to construct representations for the scarce data. Then, a new face sample is recognised by computing residuals with the second‐layer collaborative representation. Extensive experiments on four benchmark face databases demonstrate the effectiveness of their proposed approach which compares favourably against state‐of‐the‐art methods. [ABSTRACT FROM AUTHOR]

Attention priority maps are topographic representations that are used for attention selection and guidance of task-related behavior during visual processing. Previous studies have identified attention priority maps of simple artificial stimuli in multiple cortical and subcortical areas, but investigating neural correlates of priority maps of natural stimuli is complicated by the complexity of their spatial structure and the difficulty of behaviorally characterizing their priority map. To overcome these challenges, we reconstructed the topographic representations of upright/inverted face images from fMRI BOLD signals in human early visual areas primary visual cortex (V1) and the extrastriate cortex (V2 and V3) based on a voxelwise population receptive field model. We characterized the priority map behaviorally as the first saccadic eye movement pattern when subjects performed a face-matching task relative to the condition in which subjects performed a phase-scrambled face-matching task. We found that the differential first saccadic eye movement pattern between upright/inverted and scrambled faces could be predicted from the reconstructed topographic representations in V1-V3 in humans of either sex. The coupling between the reconstructed representation and the eye movement pattern increased from VI to V2/3 for the upright faces, whereas no such effect was found for the inverted faces. Moreover, face inversion modulated the coupling in V2/3, but not in VI. Our findings provide new evidence for priority maps of natural stimuli in early visual areas and extend traditional attention priority map theories by revealing another critical factor that affects priority maps in extrastriate cortex in addition to physical salience and task goal relevance: image configuration. [ABSTRACT FROM AUTHOR]

Colorization of gray-scale images relies on prior color information. Exemplar-based methods use a color image as source of such information. Then the colors of the source image are transferred to the gray-scale target image. In the literature, this transfer is mainly guided by texture descriptors. Face images usually contain few texture so that the common approaches frequently fail. In this paper, we propose a new method taking the geometric structure of the images rather their texture into account such that it is more reliable for faces. Our approach is based on image morphing and relies on the YUV color space. First, a correspondence mapping between the luminance Y channel of the color source image and the gray-scale target image is computed. This mapping is based on the time discrete metamorphosis model suggested by Berkels, Effland and Rumpf. We provide a new finite difference approach for the numerical computation of the mapping. Then, the chrominance U,V channels of the source image are transferred via this correspondence map to the target image. A possible postprocessing step by a variational model is developed to further improve the results. To keep the contrast special attention is paid to make the postprocessing unbiased. Our numerical experiments show that our morphing based approach clearly outperforms state-of-the-art methods. [ABSTRACT FROM AUTHOR]

We discuss the problem of people retrieval by means of composite pictures constructed according to descriptive portrait. An overview of the problem state-of-the-art is provided beginning from the basic concepts and terminology to a modern technology for composite picture creation, real-world scenarios and search results. The development history of systems for forming composite portraits (photo robots and sketches) and the ideas implemented in these systems are provided. The problem of automatic comparison of composite pictures with the original ones is discussed, and the reasons for unattainability of stable retrieval of originals by a composite picture in real-world scenarios are revealed. Requirements to composite pictures databases in addition to the existing benchmark databases of facial images and also methods for implementation of such databases are formulated. Approaches for generation of sketches population from an initial one that increase effectiveness of identikit-based photo image retrieval systems are proposed. The method of similarity index increasing in the couple identikit-photograph based on computation of an average identikit from the created population is provided. It is shown that such composite pictures are more similar to original portraits and their use in the discussed search problem can lead to good results. Thus the created identikits meet the requirements of the truthful scenario as take into account the possibility of incomplete information in descriptions. Results of experiments on CUHK Face Sketch and CUHK Face Sketch FERET databases and also open access identikits and corresponding photos are discussed.

Facial expression recognition (FER) plays an important role in human–computer interaction. The recent years have witnessed an increasing trend of various approaches for the FER, but these approaches usually do not consider the effect of individual differences to the recognition result. When the face images change from neutral to a certain expression, the changing information constituted of the structural characteristics and the texture information can provide rich important clues not seen in either face image. Therefore it is believed to be of great importance for machine vision. This study proposes a novel FER algorithm by exploiting the structural characteristics and the texture information hiding in the image space. Firstly, the feature points are marked by an active appearance model. Secondly, three facial features, which are feature point distance ratio coefficient, connection angle ratio coefficient and skin deformation energy parameter, are proposed to eliminate the differences among the individuals. Finally, a radial basis function neural network is utilised as the classifier for the FER. Extensive experimental results on the Cohn–Kanade database and the Beihang University (BHU) facial expression database show the significant advantages of the proposed method over the existing ones.