Scene analysis and search using local features and support vector machine for effective content-based image retrieval

• Published in: The Artificial intelligence review Vol. 52; no. 2; pp. 901 - 925
• Main Authors: Sharif, Uzma, Mehmood, Zahid, Mahmood, Toqeer, Javid, Muhammad Arshad, Rehman, Amjad, Saba, Tanzila
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 15.08.2019; Springer; Springer Nature B.V
• Subjects: Artificial Intelligence; Computer Science; Illumination; Image management; Image retrieval; Invariants; Object recognition; Repositories; Rotation; Scene analysis; Semantics; Support vector machines; Visual retrieval; Image attributes; Image matching; Clustering; Complementary features
• ISSN: 0269-2821; 1573-7462
• DOI: 10.1007/s10462-018-9636-0

Despite broad investigation in content-based image retrieval (CBIR), issue to lessen the semantic gap between high-level semantics and local attributes of the image is still an important issue. The local attributes of an image such as shape, color, and texture are not sufficient for effective CBIR. Visual similarity is a principal step in CBIR and in the baseline approach. In this article, we introduce a novel approach, which relies on the fusion of visual words of scale-invariant feature transform (SIFT) and binary robust invariant scalable keypoints (BRISK) descriptors based on the visual-bag-of-words approach. The two local feature descriptors are chosen as their fusion adds complementary improvement to CBIR. The SIFT descriptor is capable of detecting objects robustly under cluttering due to its invariance to scale, rotation, noise, and illumination variance. However, SIFT descriptor does not perform well at low illumination or poorly localized keypoints within an image. Due to this reason, the discriminative power of the SIFT descriptor is lost during the quantization process, which also reduces the performance of CBIR. However, the BRISK descriptor provides scale and rotation-invariant scale-space, high quality and adaptive performance in classification based applications. It also performs better at poorly localized keypoints along the edges of an object within an image as compared to the SIFT descriptor. The suggested approach based on the fusion of visual words achieves effective results on the Corel-1K, Corel-1.5K, Corel-5K, and Caltech-256 image repositories as equated to the feature fusion of both descriptors and latest CBIR approaches with the surplus assistances of scalability and fast indexing.