Biomedical Image Segmentation Based On Watershed And Mutual Information

Biomedical image segmentation is an important and challenging technique in image processing area. With the development of computer and microelectronic technology, large numbers of image technology have applied to the medical field, and then perfect medical image acquisition equipment system are formed which is represented by MRI, X-CT and nuclear medicine, so the medical digital image acquisition is guaranteed. The aim of medical image segmentation is to extract quantitative information of special organization in medical image, and to segregate the target in which you are interested from the background.It also made it easy, for people to analysis physiologic and pathologic information, for instance, disease diagnosis, abnormal tissue location, organ three-dimensional images and so on. Owing to the imageing characteristics of medical equipment, the diversity and complexity of human organ, it always makes medical image have acoustic noise, uneven image gray-scale, and edges are not clear, etc.In view of the above, this paper uses selective masking smoothing to remove noise and linear gray transform method to solve the problem of gray uneven. Watershed algorithm has been used to segment the biomedical image. The advantage of watershed segmentation algorithm is it can get one pixel wide, closed connected precise outline. Its shortcoming is easy to obtain over divided region. In order to suppress over-segmentation, we adopt the control marker method to mark the foreground and background respectively before watershed segment. The key of this method is to select the correct value for structure elements. Different structural elements of the segmentation may produce different results. There are few researches on how to select the best automatic structure elements to get the best effect in the literature segmentation, nobody did put forward a feasible method. This paper proposes a method for selecting optimization structure element automatically by calculating the mutual information between the original image and object marker image. Then we can get the best element value with the minimum mutual information. The performance of the proposed algorithm is evaluated by the experiment which using this paper's algorithm to segment the biomedical image. Furthermore this method can achieve good segmentation results.