The feature of isotherm in temperature indicating paints (TIP) images was analyzed

four kinds of edge extraction algorithms were experimented and compared

extracting isotherms with Canny edge extraction was suggested and the Kmeans clustering algorithm was used before the edge extraction. The Kmeans clustering and Canny edge extraction were improved. Two methods of choosing the Kmeans initial centers were proposed

which were more suitable for TIP image segmentation. The Canny algorithm to detect isotherms with higher quality was improved. The algorithm could extract clear and continuous isotherms fast. Experimental results show that the algorithm proposed runs well

and satisfies the engineering application needs. Now it has been well applied in TIP automatic temperature interpretation software development.