Color Object Recognition Using Mind Evolutionary Computation

To overcome the problems of GA, Mind Evolutionary Computation (MEC) was proposed by Chengyi Sun in 1998, which imitates two phenomena of human society -similartaxis and dissimilation. Studies have shown that the computing efficiency and the convergence rate of MEC are generally higher than those of GAs by more than 50% in the numerical optimization. Various strategies of similartaxis and dissimilation have been developed with better performance than that of basic MEC. Moreover, previous study has also shown that MEC has the ability in solving successfully the non-numerical problem, such as TSP, Job-shop and dynamic modeling of system, etc. MEChas been used in image analysis area, and attained the better results. A preliminary integrated system has already been established for MEC.Object recognition is receiving increasing attention in computer vision, and the ultimate goal of computer vision is to simulate the human perception and interpretation of the world around us. By 70s, many recognition algorithm has been proposed, and all of them can be categorized into 3 types: template-based method, feature-based method and 3D-model-based method. The principle of the template-based method is that the searched image is covered by the template image, and the template is moved up and down to look for a sub-image which is identical with the template image. The template-based method has the ability of resistance noise and translation invariance and scaling invariance, although the computation is relative more. The feature methods need to select the most effective feature from the numerous features of object, and themethod usually includes two phases: feature extraction and feature selection. Through matching the attribute of features and the relation between them, objects can be recognized. The 3D-model-based method can only recognized the known objects. It makes the model of the object and completely describes the relations between the surface, edge and position through CAD.A new object recognition method of color image is proposed in this paper, which combines the template matching with the Mind Evolutionary Computation (MEC). The recognition is done in CIE 1976 Luv uniform color space. The experimental results show that this new method has the ability to implement recognition with translation, rotation, scale and mirror invariance. Moreover, new method has the color invariance of object recognition since we recognizes the object in CIE 1976 Luv uniform color space. Furthermore, the computation efficiency of our method is high.The difficulty in object recognition is occlusion, that is, the object is occluded partly by some other objects. Because the occluded part and the occluded area are not determined before recognition, the information of feature and texture of the occluded area cannot be known.We proposed the occlusion template to solve the problem. The method is that using a circle region to occlude the template, the overlapped area between the template and the circle region is viewed as the occluded area which is marked with specifically, and the occluded area is not made the matching computation. So this kind of template is called "occluded template". The occluded template is approaching the factual occlusion area when matching.