Color Image Segmentation Based On Visual Perception Characteristics

Color image segmentation is a critical pre-process in imageprocessing. Also it’s important in the field of computer vision andpattern recognition. Image segmentation is the first step to imageprocessing in image project. The performance of the segmentationresult even may decide the quality of the subsequent image processingresult. In the last four decade, grey images are popular all over theworld. So there are many researches and algorithms about grey imageprocessing. But with the development of science and the cost down ofhardware, people can get color images everywhere more and moreconvenient. It’s a century that color images should be the definitely theleading role. Because of color images are more similar to humans’ visualperception. It contains more information than grey image. It is now veryimportant in human lives and science applications. Due to the time ofthe color image being used popular is very short. The algorithms forcolor image processing are less systematic than grey image. And manyalgorithms that used to process grey image can be used to processcolor image, too.In this paper we first state the normal model for color image. Anddescribe the physical model and the advantages and disadvantagesof each color space. Then we have a review of the development of thecolor image processing, and show how they mainly work in the colorimage process.From study of the formal algorithms, we combined the human visualand fuzzy cluster method together to segment the color image. In thispaper, we first state some evidence in the human vision research. Notall the intensity from0to255in RGB spaces can be distinguished byhuman vision. So we reduce the level of the intensity in RGB space to26,28,26respectively, while maintaining the image’s visual features and reducing the categories of image’s information. So that we canachieve the most correct peak points, from which we can form all thepossible cluster centroids. As all these cluster centroids consider theimage’s self-information, so the cluster centroids are accurate. It canreduce the times of the iterations for segmentation, so to some extentreduce the time consumption. Our experimental results show that ourmethod can be highly regulated, and the segmentation results are asgood as traditional method.At last, we put forward proposals to improve our method. And studymore about different color spaces and cluster methods to findalgorithms to segment color image more effectively.