Research On Image Super-resolution Method Based On Vector Quantization

Image is the main source that human access to information of outside world. With the rapid development of computer technology, the quality of images have been put forward higher requirements. Because of the high cost of sensors and their physical limitations of imaging units, the image super-resolution technology emerges. It can get higher resolution images without changing the image detecting system and costs less, so it received extensive concern and has been widely used in the astronomy, medicine, military and other important fields.Firstly, the thesis introduces the research background of the subject, and comments on the development of image super-resolution technology.Secondly, an image high-frequency information learning algorithm based on vector quantization and fractal thinking is proposed. The algorithm decomposes image into high-frequency information image and low-frequency information image and uses a thought which is similar to fractal and vector quantization method to learn the self-similar relationship between high-frequency information and low-frequency information. It provides a thought for image super-resolution method based on learning.Aiming at high-frequency information distortion of interpolation method which makes image fuzzy in details and contour, preserves texture poor, the thesis proposes a super-resolution algorithm based on vector quantization and interpolation. According to the self-similar relationship between high-frequency information and low-frequency information in the input image, it adds high-frequency information for the high resolution image produced by interpolation algorithms, making the output image contain more high-frequency details. Experimental results show that the algorithm performs better than traditional interpolation methods in terms of evaluations both objectively and subjectively.Finally, we design and develop an image super-resolution prototype system based on vector quantization. The system uses our method and traditional interpolation algorithm to make super-resolution processing for test images. The thesis uses the system to make evaluation and comparison of the output high resolution images.