A new implement algorithm is proposed to acquire the singularities of image based on subpixel multifractal method. The method reduces the errors of the edge measure simply generated by the gray level gradient value at the integer pixel position. The gray pixel level gradient distribution law at the position of subpixel is acquired by the algorithm combining CCD imaging mechanism. Utilizing the multifractal frame, the image is segmented into a series fractal sets of the different singularity exponents, corresponding to the image ranging from the edge to the texture. Moreover, the image edge imformation containing random details can be obtained through the method mentioned above. The influence degree of edge measure is simulated by varying unit distance of wavelet center projection. The results show that the method of subpixel dividing gradient can increase measurement robustness.Comparing most singular manifold extracted from 5×5 subpixel method and edge from Sobel operator (default threshold: 36.7920), the peak signaltonoise ratio is 9.3981 dB, when the algorithm is used in standard image segmentation. Applied on the surface cracks extraction of complex background, the results agree with the human visual reception better.