An image retrieval scheme based on block level hybrid dct-svd fused features

In this paper, an image retrieval scheme has been proposed based on block level hybrid features. The block level salient feature are extracted in two parts: first level features are formed after the application of DCT and second level features are obtained after the processing of SVD. In the first level feature, salient components are computed from image blocks based on DCT transformation, which results into DC and AC coefficients. Here, the DC component is considered as the first level feature and the AC components are processed further to get the second level feature. Now, to extract second level feature, SVD is applied over the AC components which results into singular, left singular and right singular matrices. Based on the values of left and right singular matrices, some statistical parameters are computed which serve as the second level feature for the proposed scheme. To highlight the importance of extracted feature a weight factor is assigned to both first and second level features. However, more weight is given to the significant feature i.e the first level feature than the second level feature. Also, the feature extraction process is carried out separately for all the three planes of a color image, which in return gives more detailed feature for the proposed scheme. For the retrieval mechanism, similarity is measured by utilizing five existing distance measure schemes and the results are thoroughly analyzed to check the retrieval efficiency of the proposed scheme. Due to the variable weight factor, experimental results shows decent retrieval performance and the work is comparable to the existing works in image retrieval domain.