Experimental Research on Intelligent Image Segmentation System

1指定目标区域的粗分割我们利用颜色相似性将彩色图像分割成若干区域，并将分割结果作为初始场景解释。我们使用的部分三维信息来自于初始分割的细节解释场景。通过对几种真实的场景的应用，证明了该方法的有效性。此外，我们开发了一种使用立体匹配方法获得的深度信息来解释更复杂场景的方法。2从粗糙分割中精确提取区域我们发现现有的方法不适用于由多个部分组成的物体，如人脸和衣服。然后，我们开发了一种方法，它检测到一个指定的对象和背景之间的准确边界，假设它由几个部分组成。3设计和初步实现的高速处理器的图像分割传统的高速图像处理器是不够的unage分割。在彩色图像分割中，主要采用二维直方图的构造和空间卷积（20 x ~ 20）的特征提取来提高分割速度。通过将这些模块添加到结构可变的图像处理器中，构造了高速处理器。我们通过将其应用于图像分割来检查处理时间的有效性。

1 Rough Segmentation of Specified Object RegionWe segmented a color image into several regions using color similarity, and used the result as an initial scene interpretation. We used the partial 3-D information derived from the initial segmentation to interpret the scene in details. The validity of our method was shown by applying it to several kinds of real scenes. Moreover, we developed a method of interpreting more complicated scenes using the depth information obtained by our stereo matching method.2 Accurate Region Extraction from Rough SegmentationWe found that the existing methods were not applicable to objects like humans that consist of multiple parts such as faces and clothes. Then, we developed a method which detects an accurate boundary between a specified object and the background assuming it consists of several parts.3 Design and Preliminary Implementation of High-Speed Processor for Image SegmentationConventional high-speed image processors are not sufficient for unage segmentation. Construction of 2-D histogram for initial color image segnientation and spatial convolution (2Ox2O) for feature extraction are major processing to speed up the image segmentation. A high-speed processors are constructed by adding these modules to a structure-variable image processor. We checked the effectiveness of the processing time by applying it to the image segmentation.