Vehicle Precise Retrieval via Color Image Retrieval Method Based on Improved Fast-Match

Abstract

Precise vehicle retrieval, which means ascertaining all instances for a given query vehicle image, is a challenging task as different vehicles will share the same visual attributes. To solve this problem, a novel color image retrieval method based on improved fast affine matching is proposed, which combines color constants and hue and saturation (H-S) color features of special marks to perform the fine-grained retrieval of images. The proposed method overcomes the influence of illumination variation and deformation on vehicle images. Furthermore, with full use of vehicle annual inspection labels, vehicle retrieval can be separated from excessive reliance on vehicle license plates. We evaluate our proposed method on the ReIDcar dataset, which includes many low-quality vehicle images in a series of experiments, and the large-scale image dataset VehicleID. Experimental results demonstrate that the retrieval rate outperforms other traditional methods, verifying the feasibility and effectiveness of the proposed method. Experimental results demonstrate that the proposed method is superior to other synopsis methods in the processing speed and retrieval accuracy.