Research And Implementation Of One Kind Of Key Technology For Image Segmentation

Image segmentation is an important domain in image processing area. In the existing segmentation methods, the threshold method is one of the most extensive methods in usage. Some research has been made about Maximum Between-cluster Variance in this article. The main results and achievements are as below:1. Introduce the concept of local contrast, and propose a threshold segmentation method based on 3-d maximum between-bluster variance (3-DMBV). Three dimention includes local lontrast, local average value and the pixel value. This method constructs 3D observation space using not only gray distribution information of pixels, but also relevant information of neighboring pixels. Based on the competition, redundancy and complementation of all information, it can obtain better performance than 1-DMBV method through efficient fusion. Theoretical analysis and experiments prove that it performs well on ground target image even with low SNR and low contrast.2. Introduce the concept of initial threshold, suggest to calculate the 3-DMBV in the range of initial threshold. The selection of initial threshold based on simple statistics of images, not only avoids the analysis of the complex information of images, but also avoids the optimazition of the threshold method. Use the initial threshold to reduce the time complexity, and make it suitable for the real-time application.3. Make several improvements to reduce the space complexty and raise the segmentation precision, handling with the obscure region, it not only includes all the objective and background, but also eliminates the impact of noise. These improvements reduce the solution range and increase the operation speed. Also a recursive algorithm is proposed to implement the 3-DMBV method, which can reduce the computation time and necessary storage space.This method is implemented by C++ program, and several experiments with low SNR and low contrast are used to verify the results. It demonstrates that the method which has low time complexity and space complexity can be applied to practical engineering and real-time applicaions.