Human emotion recognition based on block patterns of image and wavelet transform

In this manuscript, Human Emotion Recognition based on Block Patterns of Image and Discrete Wavelet Transform (HER-BP-DWT) is proposed. The different facial parts, such as eyebrows, eye, lips, mouth, and muscle movements play an important role in emotion recognition. But according to change in age, the movements of facial parts and muscles become weaker, so recognizing emotions can be a little more tedious and complicated. Therefore, in this manuscript, a new approach that is different from the conventional one using block patterns and discrete wavelet transform is proposed. Here, first of all, the test image is divided horizontally and vertically into different block patterns. Then, each block is separated as sub blocks. The particular area block is decomposed into different frequency sub bands with the help of discrete wavelet transform. The energy of these sub bands of each block is calculated. The energy of sub bands of the test image and the reference image is compared. The main aim of this proposed method is to recognize emotional expressions using a simple parameter, like energy of sub bands that is obtained from discrete wavelet transform and it is easy to use. The main objective is to increase the accuracy during face image recognition. The proposed HER-BP-DWT method can be efficiently and accurately recognized different emotions, such as happiness, sadness, anger, etc. The proposed method is very convenient to use due to the use of block patterns. The proposed approach is activated in MATLAB platform, then the performance is compared with other existing approaches, such as Human Emotion Recognition using Convolutional Neural Networks (HER-CNN) and Human Emotion Recognition using Bimodal Fusion Algorithm (HER-BFA). Finally, the experimental results show that the HER-BP-DWT method is superior to the existing methods. From the experimental analysis, the HER-BP-DWT method shows the accuracy of 99.55%, sensitivity of 85.93%, precision of 92.43% and 90.74% of specificity, which is prominent than the existing methods.