The goal of this paper is to present a critical survey existing literature on human and machine recognition faces. Machine faces has several applications, ranging from static matching controlled photographs as in mug shots credit card verification surveillance video images. Such applications have different constraints terms complexity processing requirements thus wide range technical challenges. Over the last 20 years researchers psychophysics, neural sciences engineering, image analysis...

Non-maximum suppression is an integral part of the object detection pipeline. First, it sorts all boxes on basis their scores. The box M with maximum score selected and other a significant overlap (using pre-defined threshold) are suppressed. This process recursively applied remaining boxes. As per design algorithm, if lies within predefined threshold, leads to miss. To this end, we propose Soft-NMS, algorithm which decays scores objects as continuous function M. Hence, no eliminated in...

Recently introduced cost-effective depth sensors coupled with the real-time skeleton estimation algorithm of Shotton et al. [16] have generated a renewed interest in skeleton-based human action recognition. Most existing approaches use either joint locations or angles to represent skeleton. In this paper, we propose new skeletal representation that explicitly models 3D geometric relationships between various body parts using rotations and translations space. Since rigid motions are members...

The past decade has witnessed a rapid proliferation of video cameras in all walks life and resulted tremendous explosion content. Several applications such as content-based annotation retrieval, highlight extraction summarization require recognition the activities occurring video. analysis human videos is an area with increasingly important consequences from security surveillance to entertainment personal archiving. challenges at various levels processing-robustness against errors low-level...

We present an algorithm for simultaneous face detection, landmarks localization, pose estimation and gender recognition using deep convolutional neural networks (CNN). The proposed method called, HyperFace, fuses the intermediate layers of a CNN separate followed by multi-task learning that operates on fused features. It exploits synergy among tasks which boosts up their individual performances. Additionally, we propose two variants HyperFace: (1) HyperFace-ResNet builds ResNet-101 model...

Adapting the classifier trained on a source domain to recognize instances from new target is an important problem that receiving recent attention. In this paper, we present one of first studies unsupervised adaptation in context object recognition, where have labeled data only (and therefore do not correspondences between categories across domains). Motivated by incremental learning, create intermediate representations two domains viewing generative subspaces (of same dimension) created...

An approach for enforcing integrability, a particular implementation of the approach, an example its application to extending existing shape-from-shading algorithm, and experimental results showing improvement that from integrability are presented. A possibly nonintegrable estimate surface slopes is represented by finite set basis functions, enforced calculating orthogonal projection onto vector subspace spanning integrable slopes. The constraint was applied iterative algorithm M.J. Brooks...

We propose a new objective function for superpixel segmentation. This consists of two components: entropy rate random walk on graph and balancing term. The favors formation compact homogeneous clusters, while the encourages clusters with similar sizes. present novel construction images show that this induces matroid - combinatorial structure generalizes concept linear independence in vector spaces. segmentation is then given by topology maximizes under constraint. By exploiting submodular...

In pattern recognition and computer vision, one is often faced with scenarios where the training data used to learn a model have different distribution from on which applied. Regardless of cause, any distributional change that occurs after learning classifier can degrade its performance at test time. Domain adaptation tries mitigate this degradation. article, we provide survey domain methods for visual recognition. We discuss merits drawbacks existing approaches identify promising avenues...

The discrimination power of various human facial features is studied and a new scheme for automatic face recognition (AFR) proposed. first part the paper focuses on linear discriminant analysis (LDA) different aspects faces in spatial as well wavelet domain. This allows objective evaluation significance visual information parts (features) identifying subject. LDA also provides us with small set that carry most relevant classification purposes. are obtained through eigenvector scatter...

Domain Adaptation is an actively researched problem in Computer Vision. In this work, we propose approach that leverages unsupervised data to bring the source and target distributions closer a learned joint feature space. We accomplish by inducing symbiotic relationship between embedding generative adversarial network. This contrast methods which use framework for realistic generation retraining deep models with such data. demonstrate strength generality of our performing experiments on...

We have collected a new face data set that will facilitate research in the problem of frontal to profile verification `in wild'. The aim this is isolate factor pose variation terms extreme poses like profile, where many features are occluded, along with other wild' variations. call Celebrities Frontal-Profile (CFP) set. find human performance on only slightly worse (94.57% accuracy) than Frontal-Frontal (96.24% accuracy). However we evaluated state-of-the-art algorithms, including Fisher...

We present an approach that incorporates appearance-adaptive models in a particle filter to realize robust visual tracking and recognition algorithms. Tracking needs modeling interframe motion appearance changes, whereas changes between frames gallery images. In conventional algorithms, the model is either fixed or rapidly changing, simply random walk with noise variance. Also, number of particles typically fixed. All these factors make tracker unstable. To stabilize tracker, we propose...

In recent years, deep neural network approaches have been widely adopted for machine learning tasks, including classification. However, they were shown to be vulnerable adversarial perturbations: carefully crafted small perturbations can cause misclassification of legitimate images. We propose Defense-GAN, a new framework leveraging the expressive capability generative models defend networks against such attacks. Defense-GAN is trained model distribution unperturbed At inference time, it...

We propose a view-based approach to recognize humans from their gait. Two different image features have been considered: the width of outer contour binarized silhouette walking person and entire binary itself. To obtain observation vector features, we employ two methods. In first method, referred as indirect approach, high-dimensional feature is transformed lower dimensional space by generating what call frame exemplar (FED) distance. The FED captures both structural dynamic traits each...

The problem of texture classification arises in several disciplines such as remote sensing, computer vision, and image analysis. In this paper we present two feature extraction methods for the textures using two-dimensional (2-D) Markov random field (MRF) models. It is assumed that given M × generated by a Gaussian MRF model. first method, least square (LS) estimates model parameters are used features. second notion sufficient statistics, it shown sample correlations over symmetric window...

Visual Domain Adaptation is a problem of immense importance in computer vision. Previous approaches showcase the inability even deep neural networks to learn informative representations across domain shift. This more severe for tasks where acquiring hand labeled data extremely hard and tedious. In this work, we focus on adapting learned by segmentation synthetic real domains. Contrary previous that use simple adversarial objective or superpixel information aid process, propose an approach...

An approach is presented for the estimation of object motion parameters based on a sequence noisy images. The problem considered that rigid body undergoing unknown rotational and translational motion. measurement data consists image coordinates two or more correspondence points. By modeling dynamics as function time, estimates model (including parameters) can be extracted from using recursive and/or batch techniques. This permits desired degree smoothing to achieved through use an...

We introduce the Single Stage Headless (SSH) face detector. Unlike two stage proposal-classification detectors, SSH detects faces in a single directly from early convolutional layers classification network. is headless. That is, it able to achieve state-of-the-art results while removing “head” of its underlying network - i.e. all fully connected VGG-16 which contains large number parameters. Additionally, instead relying on an image pyramid detect with various scales, scale-invariant by...

Some aspects of statistical inference for a class spatial-interaction models finite images are presented: primarily the simultaneous autoregressive (SAR) and conditional Markov (CM) models. Each these is characterized by set neighbors, coefficients, noise sequence specified characteristics. We concerned with two problems: estimation unknown parameters in both SAR CM choice an appropriate model from such competing Assuming Gaussian-distributed variables, we discuss maximum likelihood (ML)...

In this paper, we introduce a new synthetic aperture radar (SAR) imaging modality which can provide high-resolution map of the spatial distribution targets and terrain using significantly reduced number needed transmitted and/or received electromagnetic waveforms. This scheme, requires no hardware components allows to be compressed. It also presents many applications advantages include strong resistance countermesasures interception, much wider swaths on-board storage requirements.

We present a multi-purpose algorithm for simultaneous face detection, alignment, pose estimation, gender recognition, smile age estimation and recognition using single deep convolutional neural network (CNN). Theproposed method employs multi-task learning framework that regularizes the shared parameters of CNN builds synergy among different domains tasks. Extensive experiments show has better understanding achieves state-of-the-art result most these

An approach for restoration of gray level images degraded by a known shift invariant blur function and additive noise is presented using neural computational network. A network model used to represent possibly nonstationary image whose the simple sum neuron state variables. The procedure consists two stages: estimation parameters reconstruction images. Owing model's fault-tolerant nature computation capability, high-quality obtained this approach. practical algorithm with reduced complexity...