ABSTRACT

In coal production, it is necessary to separate waste rock from raw coal. With the development of computer vision technology, the image recognition method with simple equipment and high efficiency has become a research hotspot. Image segmentation is a very important step before the image analysis in image recognition method. Traditional image segmentation takes pixel as processing unit, without considering the influence of space and texture; therefore, it is difficult to segment the coal and waste rock images when the target and the background are with similar color and the existence of weak edges and fuzzy regions. In this paper, a new superpixel segmentation algorithm that combines color, spatial position, and texture in clustering is proposed, which integrates texture information into SLIC algorithm. After test of a great number of coal and waste rock images, the segmentation results of SLIC and SLFTIC algorithms were evaluated that SLFTIC can better fit the edge. According to the evaluation index of superpixel algorithm, undersegmentation error and boundary recall of SLFTIC are superior than SLIC, and compactness score is slightly improved. For the images with similar foreground and background and complex texture, such as mineral images and medical images, this kind of superpixel cutting method has better segmentation effect.

摘要

在煤炭生产中，必须将废石与原煤分离。随着计算机视觉技术的发展，设备简单、效率高的图像识别方法成为研究热点。图像分割是图像识别方法中图像分析之前的一个非常重要的步骤。传统的图像分割以像素为处理单位，不考虑空间和纹理的影响；因此，当目标与背景颜色相近且存在弱边缘和模糊区域时，煤矸石图像难以分割。本文提出了一种在聚类中结合颜色、空间位置和纹理的超像素分割新算法，将纹理信息整合到SLIC算法中。经过大量煤和废石图像的测试，SLIC和SLFTIC算法的分割结果被评估为SLFTIC能够更好地拟合边缘。根据超像素算法的评价指标，SLFTIC的欠分割误差和边界召回均优于SLIC，紧凑度得分略有提高。对于前景和背景相似、纹理复杂的图像，如矿物图像和医学图像，这种超像素切割方法具有更好的分割效果。