Content‐based image retrieval system via sparse representation

Abstract

The aim of image retrieval systems is to automatically assess, retrieve and represent relative images‐based user demand. However, the accuracy and speed of image retrieval are still an interesting topic of many researches. In this study, a new method based on sparse representation and iterative discrete wavelet transform has been proposed. To evaluate the applicability of the proposed feature‐based sparse representation for image retrieval technique, the precision at percent recall and average normalised modified retrieval rank are used as quantitative metrics to compare different methods. The experimental results show that the proposed method provides better performance in comparison with other methods.

A Sparse Representation Algorithm for Effective Photograph Retrieval

“…The comparison results of the precision and recall criteria are shown in Table 4. The best performance rate is P(1) = 96.66% and P(0.5) = 97.34%, which is higher than that achieved using the algorithm proposed in [25]. Table 4.…”

“…The sparse representation based method is proposed to learning dictionary via K-SVD decomposition. In [25], the iterative discrete wavelet transform is proposed to extract features and sparse representation is used to build the dictionary. Unlike these methods, we directly use the image features as a dictionary and apply the sparse vector as the similarity between the query photograph and dataset.…”

“…A combination of the precision and recall criteria was used as performance measures in [25]. The precision and recall criteria are calculated as follows:…”

“…GIST-SR 96.66% 97.34% PHOG-SR 96.16% 96.84% [25] 95.74% 96.13% Figure 6 shows some query results. If we only count the nearest retrieval result which has the largest coefficient in the sparse vector, the precision with the GIST feature is 94% and the precision with PHOG is 93%.…”

“…According to the comparison, the proposed method sees an improvement of over 10% compared with the other methods. To verify the proposed method, we have compared the proposed method with the algorithm using sparse representation [25]. The comparison results of the precision and recall criteria are shown in Table 4.…”