Research On Gray-level Image Segmentation In Complex Background

As the foundation of image analysis and image understanding, image segmentation is one of the most essential and difficult problems in computer recognition field. The existing image segmentation algorithms each have its strong point, but they are adapted to given images. Because of variety and complicacy of images, there is no uniform auto-adaptive method at present.In order to construct a method totally adapted to gray-level image segmentation in complex background, this paper summarizes and discusses common image segmentation theories and methods as well as inland and outland current research status and future development, mainly discusses threshold segmentation algorithm which is one of the most widely used algorithms nowadays.Due to varies conditions such as limitation of imaging equipment, noises in transmission and light source uniformity, a majority of collected images which have been putting into practice can be classified as images in complex background. And either color image or binary image would be transformed into grey-level image before being processed. Therefore, grey-level image segmentation in complex background is still a hot issue at the present time. Gray-level images in complex background often contain obscure target's edge, weak target-background contrast and a great many noises, a method combined surface fitting with watershed transform is proposed in this paper.Surface fitting forms threshold surface according to partial feature is a finer method but is prone to bring error. Watershed transform is efficient for partition overlapping areas but is prone to bring over-segmentation. So, this paper improves both of them. A new surface fitting method based on mean square error rule chooses the zero-cross spots detected by LoG operator to be fitting pixels is mainly presented. And a new method for separating conglutinate targets and correcting error segmentation is mainly presented too; it uses watershed transform based on distance transform to partition the surface fitting output image into exact target and the error background area. By doing this, the new method combined surface fitting with watershed transform solves the grey-level image segmentation in complex background well.In order to testify if the new method is veracious and effective, the MATLAB software is used to imitate it. Fifteen images that have characteristics of grey-level image in complex background and other ten that don't are chosen to be original input images. In comparison to the standard method using surface fitting, watershed transform, Otsu algorithm and minimum error algorithm, the new method has many advantages. The experimental results show that for the purpose of partitioning grey-level images in complex background, the new method has better capability than use surface fitting or watershed transform alone in removing background and obtaining extract target, it is established to be an excellent foundation for forward image processing.