Deep Stacked Autoencoder-Based Automatic Emotion Recognition Using an Efficient Hybrid Local Texture Descriptor

Extracting an effective facial feature representation is the critical task for an automatic expression recognition system. Local Binary Pattern (LBP) is known to be a popular texture feature for facial expression recognition. However, only a few approaches utilize the relationship between local neighborhood pixels itself. This paper presents a Hybrid Local Texture Descriptor (HLTD) which is derived from the logical fusion of Local Neighborhood XNOR Patterns (LNXP) and LBP to investigate the potential of positional pixel relationship in automatic emotion recognition. The LNXP encodes texture information based on two nearest vertical and/or horizontal neighboring pixel of the current pixel whereas LBP encodes the center pixel relationship of the neighboring pixel. After logical feature fusion, the Deep Stacked Autoencoder (DSA) is established on the CK+, MMI and KDEF-dyn dataset and the results show that the proposed HLTD based approach outperforms many of the state of art methods with an average recognition rate of 97.5% for CK+, 94.1% for MMI and 88.5% for KDEF.

A Novel Texture Descriptor: Circular Parts Local Binary Pattern

Local Binary Pattern (LBP) are considered as a classical descriptor for texture analysis, it has mostly been used in pattern recognition and computer vision applications. However, the LBP gets information from a restricted number of local neighbors which is not enough to describe texture information, and the other descriptors that get a large number of local neighbors suffer from a large dimensionality and consume much time. In this regard, we propose a novel descriptor for texture classification known as Circular Parts Local Binary Pattern (CPLBP) which is designed to enhance LBP by extending the area of neighborhood from one to a region of neighbors using polar coordinates that permit to capture more discriminating relationships that exists amongst the pixels in the local neighborhood which increase efficiency in extracting features. Firstly, the circle is divided into regions with a specific radius and angle. After that, we calculate the average gray-level value of each part. Finally, the value of the center pixel is compared with these average values. The relevance of the proposed idea is validate in databases Outex 10 and 12. A complete evaluation on benchmark data sets reveals CPLBP's high performance. CPLBP generates the score of 99.95 with SVM classification.

Composite Binary Pattern Assisted Micro-Expressions Spotting Through Feature Difference Analysis

Micro-Expressions (MEs) are one kind of facial movement which is very spontaneous and involuntary in nature. MEs are observed when a person attempts to hide or conceal the experiencing emotion in a high-stakes environment. The duration of ME is very short and approximately less than 500 milliseconds. Recognition of such kinds of expressions from lengthy video consequences to a limited Micro Expression Recognition Performance and also creates the computational burden. Hence, in this paper, we propose a new ME spotting (detection of ME frames) method based on a new texture descriptor called Composite Binary Pattern (CBP). As a pre-processing, we employ the viola jones algorithm for landmark regions detection followed by landmark points detection for facial alignment. Next, every aligned face is described through CBP and subjected to feature difference analysis followed by the threshold for ME spotting. For simulation, the REVIEW dataset is used and the performance is measured through Recall, Precision, and F-Score.

Moving Vehicle Detection in Traffic Video Using Modified SXCS-LBP Texture Descriptor

In this chapter, the authors proposed background modeling and subtraction-based methods for moving vehicle detection in traffic video using a novel texture descriptor called Modified Spatially eXtended Center Symmetric Local Binary Pattern (Modified SXCS-LBP) descriptor. The XCS-LBP texture descriptor is sensitive to noise because in order to generate binary code, the value of center pixel value is used as the threshold directly, and it does not consider temporal motion information. In order to solve this problem, this chapter proposed a novel texture descriptor called Modified SXCS-LBP descriptor for moving vehicle detection based on background modeling and subtraction. The proposed descriptor is robust against noise, illumination variation, and able to detect slow moving vehicles because it considers both spatial and temporal moving information. The evaluation is carried out using precision and recall metric, which is obtained using experiments conducted on popular dataset such as BMC dataset. The experimental result shows that the method outperforms existing methods.