Line extraction is one of the most important tasks in IC (Integrated Circuits) image processing. It is a problem of image segmentation on specialized images whose special characteristic must be taken into account during segmentation to ensure an accurate and fast result. A new algorithm of line extraction is proposed based on a detailed analysis of the similarity as well as difference between lines of IC images and that of other images. According to the unique traits of lines in IC images such as straight and direction-limited, densely populated, broken, having raise lines, the extraction algorithm is divided into four steps, i.e. a coarse extraction step by searching local color edge in limited directions, a refinement step to repair the missing parts in the coarse step by region growing, a linking step to connect, broken line regions of a common line based on their directions, a verification step to delete false lines utilizing color information. Since only steady features are used in the :algorithm, experiments show its good adaptability to different series of IC images. It can also be applied to other images when properly modified. Since color edges are selected as the local geometrical feature of lines in the algorithm above, to reduce the computation complexity and provide a proper gray image to guide the direction of gradient in color edge extraction, two adaptive color to gray transformation algorithms are proposed. From the viewpoint of dimensionality reduction, to get a gray image with the best quality, we firstly segment the sample image into regions, then maximize a modified version of fisher criterion acting as a image quality measure to get the optimal projection direction. From the viewpoint of image fusion, to enhance features and suppress noise in the fused image, utilizing the local similarity and easiness for geometrical modeling of IC images, a new degraded model for the unique local structures in IC images is constructed and the model-based fusion is divided into feature extraction, feature enhancement and noise suppression. The two algorithms are new attempts to solve the color to gray transformation problems from two viewpoints. Experiments show their superiority over traditional methods in enhancing the contrast of features of interest and suppressing noise of the obtained gray image.
修改评论