Shape as Visual Features Containing High-Level Semantic Information

Shape features containing high-level semantic information hold significant application value in content-based image retrieval and image recognition. Various descriptors can characterize shape features, with Fourier descriptors being widely applied as they simplify 2D contour information into 1D problems. However, natural image shapes are typically noisy and irregular. This study proposes an image preprocessing method to obtain purified shape images and experimentally evaluates Fourier descriptor algorithms for shape feature extraction.