Evolutionary Programming For Image Sparse Decomposition

Evolutionary Programming (EP), as the earliest branch of the evolutionary algorithm, the mutation is the only means of its reproduction, so it is simple to operate and easy-to-parallel implementation. Furthermore, EP has strong global search capability, a small number of parameters and fast convergence. But even so, EP still has some Key issues to be solved and these are very worthy of further study and discussion, for example, single mutation operator maybe result in inefficiency in the latter part of the search algorithm; sensitive to initial parameters; strong exploration but weak exploitation and so on. EP has been applied to many fields, such as the power system, tree network design, radio communication system and so on, but little reseaches show the application of EP in image processing. Therefore, EP is researched to solve the problem of computational complexity in the image sparse decomposition and to expand the application of the EP in this paper. The main work and research fruits are as follows:1. EP is used to implement the image sparse decomposition though simulation experiment. Experimental results show that EP can release the computational burden to improve the efficiency of image spare decomposition so this algorithm is the effectiveness and rationality. And then this paper reseaches and discusses three parameters of EP to provide a reference for the latter experiment in this paper.2. This paper introduces five common Evolutionary Programming algorithms, and provides the comparison and analysis of their Performance in image sparse decomposition though simulation experiment. And then, on the basis of the characteristics of EP which is applied to image sparse decomposition, a new improved Evolutionary Programming algorithm is proposed. This algorithm chooses different mutations in the different stages of image sparse decomposition and a second conditional EP is added additionally to search a better atom in a little small area around the current best atom. Experimental results show that this algorithm can accelerate the speed of convergence of image sparse decomposition and the effectiveness and rationality of this improved EP are proved.3. In order to seek the balance between exploration and exploitation, a hybrid search algorithm (IDEEP) based on EP algorithm and differential evolution (DE) is proposed, and applied to image sparse decomposition. IDEEP based fast MP is proposed to settle the issue of high computational complexity of the traditional MP approach. The fast algorithm can quickly search the best atom from over-complete dictionary of atoms, and quickly achieve the image decomposition. Experimental results show that the performance of IDEEP is better than DE and EP, so the effectiveness and superiority of the hybrid algorithm are proved.