Infrared Image Face Recognition Using Random Projection And Sparse Representation

Currently, research on face recognition has received considerable achievements, but some critical problems in this field are still needed to be resolved. One of difficulties is that the performance of face recognition system, which is based on visual light spectrum, is sensitive to the change of illumination conditions. The use of thermal infrared (IR) images can make good performance of face recognition under uncontrolled illumination conditions. But eyeglasses may result in loss of useful information around the eyes in thermal face images since glass material blocks a large portion of thermal energy, which affect the performance of recognition sharply. This paper studies robust face recognition based on random projection and sparse representation which offers the keys to addressing this problem. First the original high-dimensional of image data is projected onto a low dimensional subspace using a random projection (RP). Then sparse representation (SRC) based classification is used to resolve the eyeglasses occlusion for face recognition using infrared image, where the test image can be represented in terms of just the training images of the same object. Consider reconstituted estimation of test image computed by linear sparse structure, we propose two classification discriminators, residuals and pertinence. In order to maximize the robustness to occlusion, we introduce transverse partition scheme, apply random projection and sparse representation to each of the blocks and aggregate the results by mean value of the discrimination factor. This paper uses long-wave infrared and visible light images from the database collected by Equinox Corporation , and selects images of 44 individuals who has both eyeglasses and not. Performance evaluation experiments compare the performance of three classical algorithms, i.e., PCA, LDA, and Bayesian using both thermal infrared imagery and its corresponding visible light imagery, under improved CSU face recognition evaluation software. The results show that our algorithm is robust to the occlusion of eyeglasses in thermal infrared (IR) imagery used for face recognition and outperforms the PCA, LDA, and Bayesian algorithms.