A Study Of The Filtering Algorithm For Quasi-bilinear Interpolation Based On Curvature

With the continuous development of communication and multimedia technologies, images are more and more widely used in information transmission. However, the quality of images is inevitably affected by the internal and external interference in the process of generation, transmission, receiving and processing. This interference will not only affect the visual effect of the image that has been observed, but also prevent people from accurately understanding the information conveyed by the image. Low quality images will affect the analysis and research of image segmentation, edge detection, pattern recognition, feature extraction and bring a lot of inconvenience to the follow-up processing.According to the distribution of the noise and the characteristics of the image, people proposed various filtering algorithms. This thesis based on the characteristics of grey image and the disadvantages of the salt and pepper noise applies image interpolation technology to image processing, while it proposes a quasi-bilinear interpolation filtering algorithm based on curvature. This algorithm not only enriches the rapid development of image denoising technology, but also expands the application of interpolation technology in the field of image denoising.This thesis starts with explaining the research background, the research significance, the research status in our country and world wide and the image quality evaluation criteria. Then it not only introduces the noise types and two kinds of spatial domain filtering methods, but also discusses the image interpolation techniques and the quasi-bilinear interpolation filtering algorithm based on curvature. The paper starts from the angle of further image detail reservation, considers and refers to the curvature information in four directions, chooses the direction that is closest to the original changing regulations to have quasi-bilinear interpolation filtering. Since we consider the texture of the image, this method can preserve more information of the image while denoising. The main results of this paper are as follows:(1) According to the deficiency of bilinear interpolation in the applications of image processing, the thesis introduces quasi-bilinear interpolation filtering algorithm based on curvature. The algorithm not only considers the distance weight, also considering pixel values relative weights of the pixel to be processed and the adjacent.(2)The thesis introduces the curvature information into the image processing, and derived the formula of curvature in the digital image. The calculating method keeps the characteristics of the curvature, so that it can be better applied in the field of image processing.(3) This thesis analyses the feasibility of reference curvature in the interpolation filtering processing of the image.(4) Discussing the quasi-bilinear interpolation filtering algorithm based on curvature, at the same time, it carries on the simulation experiment and gets the ideal denoising effect. According to the interval threshold, pixels in an image are divided into two types, namely noise pixels and non-noise pixels. For non-noise pixels, the original value is reserved. As for the noise point pixels, it calculates the direction of curvature value in5x5neighborhood0°,45°,90°and135°. If four directions do not exist, the thesis searches the non-noise pixels from neighborhood and puts the median value of these non-noise pixels as the final pixel to output. If the curvature values exist only in one direction, the thesis deals with the quasi-bilinear interpolation filtering algorithm in the direction. If the curvature values of two or more directions exist, the thesis deals with the quasi-bilinear interpolation algorithm in the direction of the minimum curvature.In order to verify the validity of the algorithm, the images with low density and the images with high density are filtered under the programming environment of VC++6.0. Filtering results are compared between the median filter of image based on grey absolute relation, the improved adaptive median filtering and the standard median filter algorithm from two aspects of filtered visual effects and objective performance evaluation criteria. Simulation experiment results show the algorithm can effectively remove the noise and keep image detail to some extent. Especially in the case of larger noise density, the algorithm has better advantages than any other algorithm.