Imagery-based digital collection retrieval on Web using compact perception features

In this paper, with the objective to retrieving digital collections more intuitively and flexible, we proposed an approach of image-based digital collection retrieval on Web based on compact perception features. For this, the eigen and difference SGLD (space gray level dependence) matrices were used to extract the features of images. The associations of the extracted features with the human imagery, which can be described by the semantic, color, and structural characteristics, were analyzed. On this basis, the system of flexible image retrieval with human like performance was implemented on Web, which retrieved images by the flexible combination of query-by-sample, query-by-perception words, or query-by-impression words, and re-ranked retrieved images according to the adjustment of individual´s similarity criteria threshold. The user satisfaction-based evaluation illustrated the good performance of the proposed system.