A study for feature extraction method of color image using color histogram space and its applications

17个会计年度的主要研究工作如下：(1)基于色频信息的高精度彩色纹理图像分割方法的发展彩色纹理图像的高精度分割需要正确检测纹理边缘。本研究发展了利用颜色直方图特征从边缘存在的小区域正确检测纹理边缘的方法。如果检测到最佳纹理边缘，该技术将应用小区域中包含的颜色数成为最大值的特征。(2)利用颜色直方图改进彩色图像局部曝光的方法。本研究开发了一种校正数字图像的技术，该图像存在极亮部分和暗部分，就像在背光下拍摄的照片中实际看到的人一样。该技术利用所获取的图像自动生成一幅亮度不同的图像，并利用这两幅图像的颜色频率信息对一幅图像进行曝光补偿。该方法目前在专利申请中。(3)多光谱检测技术的发展。多光谱对光进行频率分解，可以获得传统RGB信息无法获得的颜色特征。本研究发展了我们开发的缺陷检测技术，对这种多谱特征进行了频率信息应用，实现了难以发现的缺陷。对这一多谱特征进行了频率信息应用，开发了难以发现的缺陷检测技术。

The main researches done in the 17 fiscal year are as follows.(1)Development of color texture image segmentation method with high accuracy using color frequency informationIt is necessary to detect texture edge correctly for highly precise segmentation of a color texture image. This research develops the method of detecting texture edge from the small area where edge exists correctly using the color histogram feature. This technique will have applied the character in which the color number contained in a small area becomes the maximum, if the optimal texture edge is detected.(2)Development of a method of improving local exposure in color image using a color histogramThis research develops the technology which rectifies the digital image with which an extremely bright portion and a dark portion exist like the picture taken under the backlight in a picture in which man is actually seen.This technique creates automatically one image from which brightness differs by using an acquired image, and performs exposure compensation of a image using this color frequency information of two images. This method is among patent application now.(3)Development of defect detection by using multi-spectrum spectroscopeThe multiple spectrum which carried out frequency decomposition of the light can acquire the color feature which was not acquired for the conventional RGB information.This research develops the technique of detecting the defect which we developed, which carries out frequency information application and which is hard to find it to this multiple-spectrum feature.Frequency information application is carried out to this multiple-spectrum feature, and this research develops the technique of detecting the defect which is hard to find it.