Research Of Image Recognition Method Based On Block Matching And Applications

With the development of hi-tech research and the popularity of the Internet, image recognition as a basic technology in the field of image processing is interested by many researchers. In recent years, image recognition areas involved are widespread, such as license plate recognition, fingerprint identification, face identification, precise guidance of missile, location of aircraft in remote sensing image and so on.Traditional image recognition methods which based on point pattern or line pattern perform well in general images. However, their robustness is not good when various illumination. For this issue, a matching method which based on block matching was proposed in this paper to overcome the weakness of traditional methods that suited for various illumination by way of normalized method and this way is not suitable for complex illumination. On the other hands, in general, it is necessary to denoise input images before the processing of feature extraction and matching. To join together with our proposed method, we use block matching as our starting point and doing research on image denoising algorithms firstly. Simultaneously, we intended to hold together with three main parts of image recognition:image preprocessing, image feature extraction and image matching or recognition with a clue of block matching method.Block matching method which is a means of seeking redundant information in images derived from motion estimation technology in a video sequence and serviced for some main algorithm. The typically method in image preprocessing is BM3D algorithm. The proposed algorithm achieves3D filtering in the frequency domain with the help of block matching method which makes full use of the self-similarity of image characters. However, the slightly excessive using of repeat information makes this algorithm performance not well on efficiency. Our improved algorithm is just based on this. In this paper, we attempt to simplify BM3D algorithm from the dimensions of denoising. To reducing the computational burden, we doing prejudge before the processing of3D filtering. For some flat areas in images, we think there is no need to do fine filtering, so two-dimensional filtering is chosen by us. So far, we finish the research of preprocessing in the field of image recognition by block matching. For the field of image matching or recognition, in order to simplify the problem of image matching, we first consider the case of the same scene. The great disparity is showed in images of the same scene under various illumination, that is the effect of illuminate intensity to gray values who from similar images is showed no regularity. Based on this, in this paper, we provide a hierarchical image registration method suited for various illumination. The innovation point can divide into two parts. First, in order to reduce the searching time, the whole algorithm is established in the frame of pyramid structures. Second, the correlation coefficient and structural functions are adopted to find the relations between two images. In this method, the block matching technology plays a significant role. Simulation results show that, compared with the classical full search block matching algorithm, this algorithm can reduce the time complexity effectively. Moreover, since the algorithm makes use of the details of images, it is robust to the variant illumination conditions.