Shape-Based Image Retrieval for Medical Images

Shape-based image retrieval for medical images is a crucial medical image analysis technique that utilizes shape features for efficient retrieval of similar images. In undergraduate graduation projects, this research typically involves two key phases: image preprocessing and feature extraction.

The preprocessing phase employs wavelet denoising techniques, which effectively remove noise from medical images. Wavelet transform performs multi-scale analysis to eliminate noise while preserving image edges and details, establishing a solid foundation for accurate shape feature extraction. In code implementation, this can be achieved using functions like wavedenoise() from image processing toolboxes, applying thresholding at different decomposition levels.

For feature extraction, invariant moments are selected as shape descriptors. These moments exhibit excellent geometric invariance, effectively characterizing shape features while maintaining stability under translation, rotation, and scaling transformations. By computing Hu moments or higher-order moments through functions like moments() or hu_moments(), robust feature vectors are generated for subsequent similarity comparisons. The implementation typically involves calculating seven Hu moment invariants that provide scale, rotation, and translation invariance.

During the retrieval process, the system calculates distance metrics (such as Euclidean distance or cosine similarity) between feature vectors of query images and database images. This enables shape-based similarity ranking and returns the most matching medical images. The algorithm can be implemented using distance calculation functions like pdist2() or cosine_similarity(), followed by sorting operations. This approach holds significant application value in medical image analysis tasks including tumor identification and organ classification.