Significant Spot Detection of Small Target Image in Natural Scene with Variable Neighborhood Search

Through the bright spot detection of small target image in natural scene, which is capable of appealing people's visual attention for rapid detection and identification of targets. Notable highlights model of small target image in natural scenes is prone to bring visual error with the difference between centre and surrounding of the background image, and visual saliency detection is ineffective. Considering the differences between local pixel characteristics and features of adjacent pixels, variable neighborhood search for local significance of central peripheral difference is taken, an improved significant spot detection algorithm of small target image in natural scene is proposed, constructing NSDFB frequency domain decomposition structure of significant highlight in small target image. Based on multi-scale DoG filter overlay, the calculation of the rare degree for the overall characteristics of target in the image can also measure the significance of the target, and a variable neighborhood search method is proposed to achieve improved spot detection model. The experimental results show that, the algorithm can detect bright spot of small targets in natural scene with good performance, and it has higher application value in the fields of small target recognition and multimedia information feature detection.