【作者】 黄国祥；

【导师】 汤井田；

【作者基本信息】 中南大学 ， 地球探测与信息技术， 2002， 博士

【摘要】 RGB颜色空间及其应用研究 本文初步探讨了RGB颜色空间所有颜色点转换为色度图上的颜色点的规律，按色差将所有颜色点聚类于黑蓝绿青红紫黄自8种颜色。同时，对色度图上的可见点按色度坐标剖分于这8种颜色，由此确定的剖分线与GB8416有重大区别，它适合作为RGB图像按颜色分割的一个实用标准。据此提出了色差聚类和色度剖分线2种分割彩色图像的方法。 对于彩色图像，如果按某颜色分割，那么，要增强该颜色所在波段的入射光线。对物体表面色来说，不同颜色表面的光谱反射率各不相同。彩色表面对可见光谱有较高的选择性，黑白系列的表面色对可见光谱的反射没有选择性。 对灰度直方图进行了有益的探索。(A) 给出了峰与谷的数学定义，它有别于连续函数极值的概念，也不同于用导数来定义峰与谷，而是根据给定的邻域直接比较离散数组值的大小。(B) 提出了自然段落的概念，利用它有效地简化了256个数比较结果的分类数目，有利于分类研究直方图。(C) n邻域第m点的方法可以方便地找出数组的少数几个关键点，这些关键点能较好地反映数组所示曲线的轮廓。惯穿于(A)、(B)、(C)中的思想是小波变换的多尺度思想，(A)中的邻域和(B)中的水平数就是尺度，在(C)中n就是尺度。随着邻域、水平数和n的不同，可以获得一系列不同尺度下的结果。用1个关键点来代表直方图所示曲线，当然不如用多个点的效果好，如果256个点都选为关键点，那么，关键点所绘曲线与直方图所示曲线就一模一样了。这就是逐步逼近的多尺度思想。 最后，把颜色分割的方法应用于洞庭湖水域的分割和车牌识别中，取得了较好的效果。更多还原

【Abstract】 This paper primarily discussed the law of transforming all the color points in RGB color space into color points in chromaticity diagram. And grouped all the color points into eight color such as black, blue, green, cyan, red, purple, yellow and white according to color difference. At the same time, divided the visual points of chromaticity diagram by these eight colors in term of chromaticity coordinate, then the dividing line decided is much different from GB8416. It is more suited to regard as a practical criterion for color segmentation of RGB image.To segment a color image by one color, it must enhance the incident light on the wave band of this color. As for the object’s surface color, different color surface, different spectrum reflectance. The color surface has much more selectivity, while there isn’t selectivity of visual spectrum for black and white surface color.Made valuable questioning for density-histogram. (A) Gave mathematical definition of peak and dip, it’s different from that defined by derivative and conception of extreme of continuous function as well, but compare the size of discrete array straightly based on the given horizontal value. (B) Brought forward conception of natural phrase and effectively reduced the classified numbers of 256 results by using it, so it’s of be of it to histogram’s classified study. (C) The method of the mth point in nth neighborhood can find and array’s few key points, these key points will well reflect curve’s peak and dip showed by the array. In (A), (B), (C), the multi-resolution of wavelet transformation is permeated for beginning to end. In (A) and (B), the horizontal Value is scale while n is scale in (C), when the horizontal value and n are different,a series of results of different scale will be ocguired.Of course,using a few of key points is preferable to using on key point,therefore,if the 256 points are selected,as key point,then the curve showed by then is as same as curve showed by histogram.this is the multi-resolution idea of successive approximation.Finally, applied the method of color segmentation to segmentation of Dongting-lake water area and recognition for vehicle license plates and obtained preferable effects.更多还原