Research On The Image Superresolution Technique Based On Recognition

Based on the existing imaging devices and imaging conditions,super-resolution imaging technology is to generate better quality and high-resolution images making use of single-frame or multi-frame images with poor quality and low-resolution.The key point of super-resolution imaging technology study is that how to generate high-resolution images from multi-frame (or a single frame) low-resolution images. This paper analyses the image super-resolution technology in three directions of the enlarged image sharpening processing technology, the super-resolution image processing technology based on reconstruction and the super-resolution image processing technology based on recognition, with a deep study on the latter two. The main work was accomplished as followings:1. From the three methods of the enlarged image sharpening processing, the super-resolution image processing based on reconstruction and the super-resolution image processing based on recognition, this paper investigates and analyzes the history and current situations of image super-resolution technology, analyzes the limitations of enlarged image sharpening processing technology, establishes the restricted conditions of super-resolution technology based on reconstruction, and also determines to solve the problem of resolution enhancement through the research on super-resolution technology based on recognition.2. This paper designs the limited circle neural network algorithm based on recognition, presents the idea of the algorithm and the detailed realization process. The experiments prove that the designed algorithm can effectively enhance the image resolution, and also do self-adjustment for a variety of reconstructions in different occasions through learning. The effect is the most significant when the algorithm used for processing the discrete multi-frame low-resolution images.3. This paper designs the one-way diverse learning image super-resolution algorithm based on recognition, and presents the detailed realization process of the algorithm. The experiments prove that the designed algorithm also effectively enhance the image resolution, and only needs few training samples, so the algorithm has wide practicality. The algorithm can effectively process the single frame low-resolution images, with a remarkable effect.