| Abstract: | Object recognition is a fundamental image processing problem. Due to the significant variations exhibited by diversified real world patterns, it is a challenging issue against inconsistent illumination, partial occlusion, varying viewpoint and changing background, etc. Image fusion has the merits of improving both reliability and capability for objective recognition in terms of redundancy and complementary information. However, partial information lost and aliasing impairment might occur. On the other hand, images from the human visual perspective can be enhanced by contrast stretching. It provides sensitive information through the sharp contrast changes, while the possible enhancement of noise can be resolved by adaptive algorithms and bilinear interpolations. Integration of image fusion and adaptive contrast stretching is thus proposed to enhance object recognition. Image fusion has been presented to be a primary approach of data fusion. For similar images, unique patterns occur within each individual one. The wavelet technique is studied to analyze the effect on image fusion, where 2D discrete wavelet transform is used to decompose images and reconstruct original images using the approximation, horizontal detail, vertical detail and diagonal detail components. On the other hand, adaptive contrast stretching represents a practical approach in image recognition systems. This scheme adapts to intensity distributions to improve image quality under poor visual appealing and illumination conditions. The quality of image processing should be judged on a basis of not only visual perspectives but also quantitative measures. In other words, both subjective and objective assessments should be provided. In this case, the information theory has been employed, where notions of the discrete entropy, relative entropy, gray level energy, mutual information and information redundancy are all used to indicate the impact of the proposed image processing integration approach with the 2D wavelet fusion and adaptive histogram equalization schemes. The methodology presented can be expanded to a variety of other image processing issues.
|