Human Mouth-type Recognition Based On Sparse Representation

With the rapid development of computer technology and signal processing, how to conduct image and signal processing by means of computers in a way similar to information processing by human visual perception, has become a hot research topic. Feature extraction is a key step of pattern recognition, which itself is a difficult task in computer vision. Signal sparse representation has received widespread concern in recent years. As a result of the inherent property of human visual system, the combination of sparse representation and feature extraction becomes a new research focus in the area of pattern recognition.This dissertation presents a novel scheme for human mouth-type recognition based on sparse representation. In our method, we use the K singular value decomposition algorithm (K-SVD) for training the overcomplete redundant dictionary. The dictionary consists of two sub-dictionaries, on of which is obtained by the training set of mouth-closed images, and the other, by mouth-opened images. Based on this dictionary, the test images'sparse representation is implemented by using the algorithm of Orthogonal Matching Pursuit (OMP). The sparse coefficients, namely, the extracted features, are divided into two parts, which correspond respectively to two sub-dictionaries. The two parts of sparse coefficients are employed to calculate two reconstructed images. The errors between the original test image and the two reconstructed images are respectively calculated upon which the decision of pattern identification is made. In simulation, we use the mouth-type recognition rate to verify the feasibility and efficiency of the proposed algorithm, by comparison of the result of our method with other three classical classification methods. One can conclude that the proposed technique is feasible and reliable.