Research On The Key Techniques For Pixel Level Based Multi-focus Image Fusion

Image fusion refers to an image processing technique that combines two or more source images that have been registered into a single image according to some fusion rules. The fused image obtained has more information than any single source image, which is more adapted to the human visual perception and is more suitable for the subsequent processing of the computer. Image fusion is generally divided into three levels:pixel level image fusion, feature level image fusion and decision-making level image fusion. By now, image fusion as a very important technology has been widely used in computer vision, medicine, remote sensing, military, public security and so on.Due to the limited depth of field of optical lenses in CCD devices, it is often impossible to get an image that contains all relevant objects in focus, which means if one object in the scene is in focus, another one will be out of focus. Therefore, in order to obtain all information of the scene, images taken from the same scene focused on different objects need to be fused, that is multi-focus image fusion. After fusing, the resultant image not only contains the redundant information and the complementary information of each individual source images but also contains information that can not be embodied in any single source images. Obtaining an image that contains all relevant objects from the same scene in focus, which is more suitable for the subsequent processing of image, such as feature extraction, image segmentation, target detection and recognition and so on.In the paper, serious problems of the multi-focus image fusion algorithms at pixel level are studied, mainly including the following aspects:Firstly, to overcome the some problems that the conventional block-based multi-focus image fusion algorithms faced with about how to determine the size of the sub-block and which one index is to be chosen as the evaluation criterion to measure the sharpness of the sub-blocks, a novel automatic block-based multi-focus image fusion with genetic algorithm (GA) is proposed in the paper, in which the block size can be automatically found. In the method, the size of the sub-block is handled as a chromosome, and Sum-modified-Laplacian (SML) is selected as an evaluation criterion to measure the clarity of the image sub-block, and the edge information retention is taken as a fitness function to calculate the fitness of each individual. Then, through the selection, crossover and mutation procedures of the GA, we can obtain the optimal solution for the sub-block, which is finally used to fuse the images. The experiments show that the proposed algorithm outperforms a series of traditional multi-focus fusion algorithms and two existing GA based multi-focus fusion algorithms in the literatures.Secondly, up to now, most of the multi-focus image fusion algorithms use single feature to measure the pixel sharpness. Measuring the pixel sharpness with single feature is not comprehensive enough. Moreover, almost single features don’t take into account the human visual perception characteristics. Considering the characteristics of artificial neural network with multiple inputs and the human visual perception characteristics, a novel pixel level multi-focus image fusion algorithm based on human visual perception characteristics and BP neural network is proposed in the paper and compared with other traditional multi-focus fusion algorithms. The experiments show that the performance of the proposed algorithm is superior to several existing fusion algorithms.Thirdly, to overcome the shortcoming of the traditional fusion algorithms in transform domain that use single feature to measure the pixel sharpness and consider the characteristics of the non-subsampled contourlet transform (NSCT) coefficients and human visual perception characteristics, a novel NSCT-based multi-focus image fusion algorithm is presented in the paper. Compared with series of fusion algorithms, including wavelet transform, contourlet transform (CT), NSCT, the existing NSCT based fusion algorithm, and the proposed algorithm can get better visual effect and higher values of edge information retention and mutual information, which verify the superiority of the algorithm.