Fusing Joint Multiplexed Curvature Gradient and Color Features for Image Retrieval
TL;DR
This paper presents a new image descriptor for image retrieval, which aggregates color information and multiplexed curvature information, which outperforms sophisticated algorithms and achieves excellent efficiency.
Abstract
Content-based image retrieval has become a popular research area, and image retrieval algorithms using the multi-channel mechanism or multi-feature extraction strategy usually can achieve competitive results. However, the robustness against rotation and scale variation is unsatisfactory in those algorithms. This paper presents a new image descriptor for image retrieval, which aggregates color information and multiplexed curvature information. In this work, different sampling windows are utilized to capture curvature gradients to enhance the performance of the image descriptor. Specifically, a large window is employed to capture curvature gradient information that is multiplexed and integrated with gradient features from other scale spaces. Simultaneously, a mask mechanism is adopted to reduce redundant information. Next, the color histogram is combined to increase the robustness of the algorithm. Finally, comparative experiments conducted on three datasets demonstrate that our algorithm outperforms sophisticated algorithms and achieves excellent efficiency.