A Gray Level Image Detection and Segmentation Algorithm Based on Average Intensity Project and Shift Gaussian Model

Abstract

A gray level image segmentation algorithm is proposed based on Average Intensity Project (AIP) and shift Gaussian model for gray level images with connection between thread and area with similar gray level. The evaluation is performed on 30 gray level images, and the results show area overlap measure of 91% between the extracted characteristic parameter and the manual segmentation. Introduction Gray level image segmentation algorithm is one of main support to judge the edge parameter which be used to recognize the great value target. Due to some target is very similar in pixel gray with its surrounding environment; therefore, it is an emphases and a difficulty that research to how to divide exactly vascular adhesion area. Now, mainly methods to divide similar gray level image are these: utilizing morphological method to divide vascular adhesion in literature [1], but the parameter of canker bulgy is to be control difficulty, so make for lack or exorbitance, and probably eliminate its edge burrs. In literature [2], to use with these step that threshold segmentation, model setting, and nodule segmentation, due to modeling pertinence is weak relatively, so its effect is not good. In literature [3], an amelioration method on C equal-value clustering is used to divide gray lever image, the characteristic parameter can be picked up, but this method can be used to divide 2 or more. In literature [4], it rebuilt three-dimensional image in utilizing planar CT image, it can be revert characteristic parameter of the area, but its computation is very most, and can not solve effectively the problem in great area, or its borderline is not clear. The reason which not to pick up characteristic parameter exactly is, that the information of gray lever images is not use sufficiently, other said that module building is not based on point of pixel gray and geometry characteristic, so some information on originality image is lost.