Research Of Color Image Segmentation Method And Its Application Based On Visual Features

With the development of capturing equipment, the accuracy and stability of computer vision system needs to be improved for the human's requirement. Color image segmentation has become a basic and critical step in computer vision system. How to segment the high-resolution color image fast and accurately, this work owns not only great academic value but also great commercial value.This paper takes color image as its subject, has done further researches on watershed segmentation algorithm and graph cut segmentation algorithm, achieved the following results:(1) As to the problem that traditional watershed transform algorithm always generates excessive numbers of small regions. This paper presents a new watershed pre-segmentation algorithm based on the smoothed original image and its smoothed gradient image. The new algorithm smoothes the original image first, construct the gradient image from the smoothed original image, then smoothed the gradient image again, and the following steps are same as the traditional algorithm. This is just a Pre-segmentation algorithm. Experiments show that the presented algorithm not only maintains the image's local information perfectly but also reduces the number of small regions generated by watershed transform effectively.(2) As to the problem that the method based on graph cut takes too long time when processing high-resolution color image, which cannot meet the real-time processing requirement. This paper presents a fast graph cut segmentation algorithm based on the image pyramid model. The presented algorithm coverts the high-resolution color image to low-resolution color image firstly, then performs graph cut algorithm on the low-resolution color image, then propagates the segmentation result to the high one, by the help of the low levels segmented result we could get the final result at last. Under the same experimental environment, Experiments show that the method presented in this paper could accelerate the image segmentation speed and reduce the memory consumption as the low-resolution image reduced the number of graph vertex.(3) In order to improve segmentation accuracy and reduce segmentation time. This paper uses the small regions generated by watershed transform to construct the graph, and assigns the similarity between the small regions to the graph edge weights, finally uses the maximum flow-minimum cut theorem to complete the image segmentation. Experimental results show that the new method maintains the segmentation accuracy and reduces the segmentation time effectively.(4) Based on the above results, we design an interactive image segmentation system, which needs less human interaction, easy to operate and segment the color image fast. The idea of the interactive system has been successfully applied to the Znse Optimized Cutting system based on computer vision, with small amount of human interaction. The computer vision system could segment the low quality color image captured because of Znse materials reflecting and the complicated background of the package. This has improved the accuracy, stability and efficiency of the computer vision system.