Identification Of Static Facial Expressions Of People Based On Sparse Representation

With the popularity of computers and Internet, the relationship between humans and computers becomes closer, and with a broad application prospectes, facial expression recognition has become one of central issue in science step by step. However, the diversity of facial expression brings a great problem to study, so the key point of this paper is how to correctly identify and distinguish what type the facial expression belongs to.Due to the complexity of facial expression research, this article mainly illustrates by the following three aspects:image pre-processing, feature extraction and feature classification. The main task is as follows:1. The part of image pre-processing. In this part, the face image was preprocessed, mainly through size normalization and intensity normalization. At the same time, the focus is the division of the image block, and different weights for different sub-regions are given.2. The part of feature extraction. In this part, discussed several methods for sparse solving and selected the most suitable method. And in the facial expression recognition, common feature extraction methods of Gabor, LBP operator and CBP operator, then CTP operator is put forward. By comparing the former, the method which fusion Gabor and CBP operator is put forward, then integrate the sparse representation with histogram dimensionality reduction and extract facial expression features finally. This method proved by experiments indicates that the effects are significant.3. The part of classifier design. In this part, several conventional classifiers were discussed simply, and focus on the nearest classifier and the expanding------KNN classifier and center nearest neighbor classifier. And a detailed analysis of their principles and performance in facial expression recognition is conducted. The paper proposed classifier design based on the fusion of sparse representation and the nearest classifier and by calculating the minimization and the minimum residual difference to determine their type.4. The part of experimental verification. In this part, through experimental comparison and selected the most suitable method. Then based on the experiment to prove the better result of the new method which is refer to in the paper. At the same time, in this article, it raised self-building expression database for verifying the improved method, which effectively proves the excellent robustness of the improved method.