Human Age Estimation Method Robust to Camera Sensor and/or Face Movement

Abstract

Human age can be employed in many useful real-life applications, such as customer service systems, automatic vending machines, entertainment, etc. In order to obtain age information, image-based age estimation systems have been developed using information from the human face. However, limitations exist for current age estimation systems because of the various factors of camera motion and optical blurring, facial expressions, gender, etc. Motion blurring can usually be presented on face images by the movement of the camera sensor and/or the movement of the face during image acquisition. Therefore, the facial feature in captured images can be transformed according to the amount of motion, which causes performance degradation of age estimation systems. In this paper, the problem caused by motion blurring is addressed and its solution is proposed in order to make age estimation systems robust to the effects of motion blurring. Experiment results show that our method is more efficient for enhancing age estimation performance compared with systems that do not employ our method.

Person Recognition System Based on a Combination of Body Images from Visible Light and Thermal Cameras

“…As proven through various previous studies [7,44,45,46], this method offers the illumination and rotation invariance characteristics for the extracted image features. As a result, this method has been successfully used in many image processing systems such as face recognition [7], gender recognition [44], and age estimation [45,46]. Mathematically, the LBP method extracts a descriptor for each pixel in an image using Equation (1).…”

“…From the classified uniform and non-uniform patterns, the image feature vector is formed by accumulating a histogram of uniform and non-uniform patterns over the entire image. Inspired by the research of Nguyen et al [46] and Lee et al [7], we use multi-level local binary pattern (MLBP) to extract the image features of a given image. The difference between LBP and MLBP is that the MLBP features are obtained by dividing the image into sub-blocks with different sub-block sizes and concatenating the LBP features of all sub-blocks together to form the MLBP feature.…”

“…The difference between LBP and MLBP is that the MLBP features are obtained by dividing the image into sub-blocks with different sub-block sizes and concatenating the LBP features of all sub-blocks together to form the MLBP feature. Consequently, the MLBP features can capture more rich texture information (both the local and global texture features) than the LBP features [7,45,46]. ${\mathrm{LBP}}_{R,P}={smalltrue\sum }_{i=0}^{P-1}sfalse(g_i-g_{c}false)\times 2^i\text{ where }\mathrm{s}false(\mathrm{normalx}false)=true\{\begin{array}{c}1, if x\ge 0\\ 0, if x<0\end{array}$…”

Gender Recognition from Human-Body Images Using Visible-Light and Thermal Camera Videos Based on a Convolutional Neural Network for Image Feature Extraction

“…Therefore, the extractors simply capture several aspects of the problem. For example, the LBP method is designed to count the number of uniform and non-uniform image texture features in an image [5,18], the HOG method is designed to capture the edges and edge strengths [5,10,12,19], and the BIFs method is designed to extract the image features using different bandwidth and texture direction using Gabor filters [11]. Given the predesigning approach, these image feature extractors have fixed parameters and definitions even though they are applied to various types of images and/or different image textures.…”

Motion Deblurring of Faces

“…The domain-specific methods in face deblurring can be classified in two categories: i) those that include joint optimization for solving another task (Liao et al (2016); Nguyen et al (2015); Ding and Tao (2017)), ii) those that utilize the geometric information, e.g. shape/contour (Pan et al (2014a)).…”