The Research On Multi-scale Image Structure Enhancement Algorithms And Application

The medical and industrial images often suffer from the problems of poor contrast and noises. The enhancing algorithm for the images could significantly improve the quality of the images. In these images, there are many local features, which keep a certain direction and size, and the gray level across the profiles often takes a symmetrical distribution. The scale-space based derivative operator is an effective way to extract the local features. This paper studies the image feature detecting techniques via the design and implementations of two examples.This paper reviews the background and status of the multi-scale theory and the local feature extracting methods, presents two examples based on a detailed explanation of the scale-space properties, selection of the scale and several major local feature extracting methods. The two examples in the paper refer to two medical and industrial image fields, which are image enhancement filtering of fissure and cigarette cases’ quantity detecting algorithm:1) In the design of image enhancement filtering of fissure, we presume the local geometrical structures of fissure belong to surface-shape. In the framework of multi-scale, we decide the local geometrical structure of each pixel via the analysis of Hessian matrix eigenvalues and enhance the pixels in the surface-shape structure while restrain the one in other geometrical structures. Lastly, the filtering achieves a proper result in the experiment of auto-enhance a pulmonary CT image.2) In the design of cigarette cases’ quantity detecting algorithm, we first collect the one-dimensional signals that captured by an industrial CCD camera from the side-view of cigarette cases, then process the signals by multi-scale Gaussian differential operator filtering, crest detection, adaptive threshold and so on. The process of multi-scale Gaussian differential operator filtering is implemented by the comparison of Gaussian second differential operators and Bi-Gaussian operators. In the end, we conduct a large amount of experiments and reach a conclusion that there is no miscalculation when the sides of cigarette cases do not have serious scratches.