Based On The Multi-scale Image Decomposition Of The Image Filter

Image Enhancement has always been a hot issue in the realm of computer vision. Research topics on Image Enhancement emerge in endlessly and simultaneously as the realm of computer vision is growing wider and wider. Image Enhancement research which involves several subjects such as Biomedicine, Artificial Intelligence, Pattern Recognition, Electrical Engineering and so on is to achieve the image adjustment with the use of Basic Math and computer theory, thus accommodating diverse application requirements. Current technology grants Image Enhancement more application directions despite its relative new appearance. Image Enhancement has extended from simple Image Denoising and image details exaggeration to the leading edge like High dynamic range Image Tone Mapping, Image Defogging Effect, Image Abstraction etc.Former filters in Image Enhancement usually have several limitations. For example, though linear filters have a good performance in image denoising, it can only be achieved at the cost of the edge of images. On the other hand, nonlinear filters succeed in preserving the edge, but it need more complicated calculation. What’s more, when it comes to stronger texture images, nonlinear filters are not able to smooth texture images and save edge at the same time. On the ground of existing algorithm of image filters, this paper analyzed the characteristics of both linear and nonlinear filters, raised several modifications according to the major design direction of these two filters—pixelbypixel and optimization algorithm. In the same time, this paper also designed an image filter which can be able to both extract texture information and save edge. The contribution of this paper may go as follows:1. Analysis of two design directions of current two image filters and characteristics of diverse designs. Several modifications of filters have been raised in order to get better performance in other realms of image research.2. A new image filter has been designed to be able to both extract texture information and save edge on account of overcome weakness in preserving edge of current filters. This method also can perform well in High dynamic range Image Tone Mapping and image details exaggeration. With the experimental result and performance analysis, the improvement of current algorithm and the new algorithm both raised by this paper can be seen to perfectly meet different needs of image details extraction. In addition, these methods can also be applied to other realms such as High dynamic range Image Tone Mapping, image details exaggeration, Image Abstraction etc.