Automatic facial expression recognition has always been a challenging task to understand human behavior from real world images. Certain type of issues are associated with such images that include poor illumination, di...
详细信息
Automatic facial expression recognition has always been a challenging task to understand human behavior from real world images. Certain type of issues are associated with such images that include poor illumination, different orientations and varying pose. The proposed technique first applies Fast Fourier Transform and Contrast Limited Adaptive Histogram Equalization (FFT+ CLAHE) method to compensate the poor illumination. Then mergedbinarypattern code (MBPC) is generated for every pixel. Two bits per neighbourhood are produced to form a 16-bit code per pixel. This code merges local features to improve the effectiveness of facial expression recognition system. MBPC descriptor captures changes along fine edges and prominent pattern around eyes, eye brows, mouth, bulges and wrinkles of the face. The results of proposed technique are compared with different variants of LBP and LGC based techniques for both holistic and zoned images. Static Facial Expression in Wild (SFEW) dataset is selected for experimentation. Results clearly indicate that the suggested MBPC based technique surpasses other techniques with 96.5% and 67.2% accuracy for holistic and division based approach respectively. Moreover, results indicate that the performance of holistic approach is much higher than division based approach. (C) 2018 Elsevier GmbH. All rights reserved.
暂无评论