基于多尺度的纺织印染图像分色算法研究

Research on color separation algorithm of textile printing and dyeing image based on multi-scale

对基于上下文信息的多尺度纺织印染图像分色算法的效果进行研究。比较上下多尺度图像分色算法与Mean-shift分色算法的分色结果与边缘轮廓的提取效果,发现多尺度分色算法的效果比Mean-shift分色算法的效果好。实验结果显示,多尺度分色算法利用多尺度上下文模型对分色结果进行不断修正,通过人眼观察设定图案的主要色调并且通过RGB空间与HSL空间变换,计算图案区域的色度值,进行纹理噪声及边缘孤立区域的后处理。比Mean-shift分色算法计算出的主要色调更加符合原始印染图案,并且避免了纹理噪声的干扰。

The effect of multi-scale textile printing and dyeing image color separation algorithm based on context information was mainly studied. The color separation results and the extraction effect of edge contour of the multi-scale image segmentation algorithm and the Mean-shift color separation algorithm were compared. It was found that the effectiveness of multi-scale color separation algorithm was better than that of Mean-shift color separation algorithm. The multi-scale color separation algorithm used multi-scale context model to modify the color segmentation result continuously. The color values of the pattern region were calculated by observing the main hue of the pattern set by human eye, and the chromaticity value of the pattern region was calculated through the transformation of RGB space and HSL space, then the post processing of the texture noise and the edge isolated region was carried out. Compared with the Mean-shift color separation algorithm, the main hue calculated by the multi-scale image color segmentation algorithm was more consistent with the original printing pattern and avoided the interference of the texture noise.