A Fast Specific Object Recognition Algorithm in a Cluttered Scene

Abstract

Specific object recognition technology is an important research component of computer vision and image processing technology and is also used in Industrial Internet and Internet of Things. In recent years, due to the widespread use of visual surveillance systems, specific object recognition technology has been gradually applied in monitoring systems based on image processing. By satisfying image feature invariance under changes in such factors as scale, illumination, and rotation, point-matching methods based on local invariant feature transform (LIFT) have recently become an attractive field for specific object recognition. In this paper, we propose a fast specific object recognition algorithm in a cluttered scene based on LIFT. First, the pyramid level of the reference image that is closest to the scale of the object in the scene image is determined and is referred to as the corresponding level. Next, the resolution of the scene image is increased by enlarging the interpolation based on the corresponding level. Finally, LIFT matches the reference image and interpolated scene image on a single level, reducing false match pairs in addition to considering the number of keypoint restriction problems. The experimental results demonstrate that the proposed algorithm is not only fast but also highly robust compared to existing algorithms. Thus, it can recognize objects correctly.