A wavelet-PCA approach for content-based image retrieval

This work describes a novel and efficient algorithm for content-based image retrieval based on discrete wavelet transform (DWT) and principal component analysis (PCA), together with inputs drawn from Euclidian operator, a common criterion to measure distance among matrices. The former is used to produce a signature from the query input image, a compressed and codified matrix that holds the key features of the original data, and the latter is used to obtain the projections of the original data onto particular subspaces. Interestingly, the tests state that for each particular query, the worse the frequency response of the analysis-filter used is, the better the classification is, 98.61 % being the best accuracy the algorithm has reached. The system's input consists of a query image and its output corresponds to the most similar image found in the data-base, according to the distance criterion adopted.