Ear Identification Based On Sparse Representation

The human ear recognition is an important technology in modern biological pattern recognition, and is a kind of technology identifying the person in an image through a given image to the human ear by using the stored ear database. Most of the existing identification methods by using characteristics of human ears require image preprocessing or complex feature extraction. However, choosing what kind of characteristic and even the descending dimension method selected will have a great influence on the final recognition rate. Meanwhile, in practical application, human ear image identification under interference from sheltering, noise, attitude rate etc. has been one of the difficulties in the research field.Sparse Representation-based Classification is the successful application of compression perception theory in pattern recognition. Compared to the traditional corner detection algorithm, etc., sparse representation has a strong robustness in pattern recognition for sheltering, noise, attitude rate. This paper conducted a detailed study of the human ear recognition specifically for ear recognition with multiple attitude changes. The works and achievements are as follows:(1)Conducted systematic studies and researches on ear image preprocessing methods and performed processes for human ear images such as gray scale, scale normalization, lighting effect removal, filtering enhancement etc., which make the processed images more suitable for the study of the human ear recognition.(2)Conducted studies and researches on sparse representation theory, illustrated the existence of sparse solutions, the optimal solution of the sparse solution method, the concrete steps and its limitations of sparse classification, experiment verified the feasibility of ear recognition with sparse classifier, and analyzed the causes and improving directions for identification error.(3)Specifically for the over-complete of a optimization dictionary, selected the HOG feature dictionary, proposed a human ear recognition method based on HOG sparse representation, and verified the improvement and feasibility of the proposed method by contrast experiments and human ear recognition experiment with multiple attitude changes.(4)From the perspective of sparsity of optimal weight coefficient, selected OMP algorithm for sparse solution with L0 constraint, proposed a human ear recognition method approximate to sparse solution based on L0 sparse constraint, and performed ear recognition through the biggest weight coefficient of sparse solution and verified the improvement and feasibility of the proposed method by contrast experiments.