Research And Application On Image Inpainting Based On Texture Synthesis And Image Segmentation Based On Fractal

With the development of electronic technology and computer technology, digital image technologies are widely used in the field of digital television, digital video camcorders, digital cameras and other digital products. People's need for digital image processing is increasing, In order to meet people's need, to achieve a definite purposes, people pay more attention to the research of the special treatment of the image.These special processing includes to repairing the damaged parts of images, it makes the restored image close to or reach to the effects original image, it is called as image inpainting. It also includes the technology which can separate the target and background, it is called image segmentation.Image inpainting is to restore the missing or damaged portions of the image in order to make it more legible and to restore its unity in a way that is non-detectable for an observer who does not know the original image. Currently, digital inpainting techniques have found broad applications in image processing, vision analysis, digital restoration of ancient paintings for conservation purposes, text removal and objects removal in images for special effects, restoration of old photographs or films with scratches or ancient paintings for conservation purposes, text removal and objects removal in images for special effects, restoration of old photographs or films with scratches or missing patches, occlusion in computer vision, errors conceal in videos, and so on. Therefore people pay more attention on image inpainting, digital image inpainting technique is becoming an important subject of research academia.In the image processing and image analysis research and application areas, image segmentation is a fundamental and key technology. Image segmentation is to separate the target and background, on this basis, it is possible to further analysis and use of targets, it provides an important basis for the follow-up treatments. The results of its analysis or processing will directly affect the subsequent processing of information. So we say that image segmentation is an important part and key steps from image processing to image analysis. Therefore, study image inpainting and image segmentation is so significance to further development of the theory for digital image processing and further expand the application field of image processing.Rely on the item of 863, this paper mainly studies on image inpainting and image segmentation based on fractal. For image inpainting, this paper pays more attention to the image inpainting algorithm based on texture synthesis. For image segmentation, it focuses on the image segmention algorithm based on fractal dimension. Finally, the results of this study will be applied to the system of the cloud spatial distribution and orientation, it gets good results.The main work and innovations are as follows:(1)This paper presents a method of segmenting the pre-selected regions in the images to be repaired, this method is defined by simple mathematical expression, divided pre-selected regions from the target areas and then search a best-exemplar from pre-selected region to inpainting the image by MRF. The target areas is completely determined by the pre-selected regions to some extents. This method prevents the block wrongly matching effectively, it can obtain good inpainting result in quality.(2)This paper presents the adjustment rule of adaptive window size. This adjustment rule is definite according to the changes of image gradient can indirectly reflect the changes in image spatial frequency, it can determine the size of the repair window automatically according to the level of the image frequency and the amount of information contained. This method improves the repair quality effectively.(3) This paper presents the method of downscale original image by bilinear interpolation, after that inpainting the image in the shrinking images. This method reduces the recovery time significantly, improve the effectiveness of the algorithm greatly.On this basis, this paper presents a novel method of efficient image inpainting method based on region segmentation and varying exemplar. Firstly, shrinking the original image to 0.02-0.25 times through the downscaling method, and segmenting the pre-selected regions in the downscaled image as the source region, using the adjustment rule of adaptive window size to determining the fixed window size; Secondly, search a best-exemplar from pre-selected region to inpainting the image, then using the method of image segmentation to inpainting the regions of incomplete restoration in the downscaled image. Finally, these are filled into the inpainted region of the original image. Iterate the above steps until the whole image inpainting is completed. The result demonstrates that this method is 100～5 times the efficiency of the method which is existed now.It can obtain good inpainting result in efficiency and quality.For image segmentation, this paper first analyzes the traditional algorithms based on fractal dimension, an efficient image segmentation method based on local fractal is proposed, through analyzing the factors which affect the efficiency of the algorithm. Firstly, utilizing an array as a data structure, fractal dimension can be calculated fast, and then identifying the fractal dimension exceeds a certain threshold area as the clouds of regional by analyzing the fractal clouds. Finally, each horizontal line of high-dimensional is calculated to get the segmentation results. The result demonstrates that this method is 6-80 times the efficiency of the method which is existed now, and then distinguishing the clouds, and other artificial objects effective. It can obtain good segmentation result in efficiency and quality.Finally, the results of this study will be applied to the system of the cloud spatial distribution and orientation, image inpainting method is utilized to remove the interference of camera bracket, image segmentation method is utilized to extract infrared cloud image, after that pseudo-color processing is utilized. The location information of cloud distribution is determined by establishing the relationship between the pitching angle and the radius on the image. Experiment results show that the method utilized in this paper is simple and effective, the interference problem existed in the infrared image is solved, the cloud spatial distribution and orientation is achieved. This method can also obtain good result in efficiency.