Study on non-destructive extraction methods of ancient ceramic cultural elements ornamentation

Aiming at the existence of ancient ceramic ornamentation with complex ornamentation and lack of information on ornamentation, there is a problem of noise in the image. In this paper, a method for extracting features of ancient ceramic ornamentation is proposed. The method is based on dynamic truncated mean filtering and double-thresholding the maximum interclass variance. Second-order derivative filters were constructed to compute the second-order derivatives in the local x and y directions for each pixel of the image, a sigmoid function was constructed to map the curvature values into the [0, 1] region, and the size of the filtering window is adaptively adjusted according to the local curvature of the image, for better preservation of edge information while smoothing the image. Combined with double-threshold maximum interclass variance, the image is binarized by two thresholds being selected, and the image with high interclass variance is selected as the final segmentation image. Experimental results indicate that the filtered image PSNR of this paper’s method is 32.4720, the precision of the pattern extraction reached 0.910 15, and the effects of noise and ornament complexity on the accurate extraction of ornamentation are effectively addressed.