M. SreeRajeswari and M. Rajalakshmi. Feature Extraction Methods and Re-Ranking with Click-Based Similarity for Web Images. Dynamic Systems and Applications 30 (2021) No.6, 944-963

https://doi.org/10.46719/dsa20213063

ABSTRACT.
Web-scale image search engines mostly depend on surrounding text features. It is difficult for them for user’s intention prediction only with query they are giving and this leads to ambiguous and noisy search results from the search engines which are far from satisfactory. In this research work, the proposed approach performs system on web image retrieval by implementing a system of searching through word-based as well as features and multiple features with different modalities. Correlating terms along images, textures and color in web are performed with modalities accordingly. Color feature extraction is performed by using the RGB, YUV model. In sequence, co-occurrence matrix is performed by extraction of texture feature. Then Click-based Multi-feature Similarity Learning (CMSL) performs re-ranking organization that follows the feature extraction. Whereas the conduct Incremental Spectral Clustering (ISC) for clustering visually and semantically alike images into similar group. At last, Multiple Support Vector Machine with Kernel Learning (MSVM-KL) is proposed for combing modalities of several visually holds distinct as well as integrated similarity space. The final list of re-ranked are attained along computation of all images in clusters images by determining the local density and click-based initial image confidence.

KEYWORDS: Keyword-Based Search, Color Feature, Texture Feature, CMSL, MSVM, Kernel, Re-Ranking, RGB, YUV, Click Through Data.