Performance analysis of canny edge detection for illumination invariant Facial Expression recognition

Abstract

Face perception is a very important component of human cognition. We can judge the person's mood and mental status through his/her expressions. In other words, the most expressive way human display emotion is through facial expressions. And hence facial expression recognition has become an active research area in the field of human computer interaction. The work in this paper concentrates on images having different illuminations and analyzes the performance of canny edge detection method with two classifiers, Euclidian distance and neural network. The results are tested on JAFFE (Japanese Female Facial Expression) database, available in public domain and IFE (Indian Facial Expression) database which is self created.

Facial expression recognition based on regional adaptive correlation

“…The authors in ref. [9] proposed a covariance pool structural based on the VGG network and used it as a regional descriptor. The authors in ref.…”

“…There are many research studies using the second‐order statistics to describe structural information. The difference between our work and other works [3, 9, 18–22] is as follows: (1) The model proposed in this paper does not use manifold network calculations and is trained end‐to‐end based on FC layers and Softmax for classification. (2) This paper does not directly use the features of the second‐order statistics as the structural features, but first uses the second‐order statistics to encode the CNN features of the extracted local regions, and then select the more discriminative structural features adaptively through the designed algorithm.…”

“…The first-order features are computed on the original data, and the higher-order features are computed on the original data or the first-order feature descriptors. CNN that typically uses convolution layers, maximum or average pooling layers, and FC layers is considered to capture only the first-order statistical information [9]. Recent works [10,11] have shown that second-order statistics (information) in CNN can effectively model spatial and channel correlations between features and mine the useful information contained in the features themselves.…”

Illumination Effects on Facial Expression Recognition using Empirical Mode Decomposition

“…In previous work, limited study was conducted on illuminations effects for facial expression recognition. For example, [3] have combined the logarithm transform, discrete cosine transform, and illumination compensation as a normalised DCT to eliminate illumination variation contained in the facial images of the JAFFE database. To recognise the facial expression, the extracted features from the pre-processed image were combined with a neural network classifier.…”