Analyzing optical texture of metallurgical cokes is an important way to measure the quality of coke.Currently,these measurements have been still in the manual stage.In this paper,we present a system to improve the status quo,which can automatically capture the microscopic images of coke sections and then segment and recognize different coke optical textures in the images.The system consists of two subsystems,which are image acquisition system and image analysis system.We take advantage of multi-directional polarizing technique to capture the reflectance images of coke sections at different polarizing angles,and this special type of image is termed as hyper-reflectance image in this paper.In the image analysis system,we propose a novel analysis algorithm to analyze the hyper-reflectance image of coke,which shows excellent performance in the segmenting and recognizing of coke optical texture.