Image specific target detection and localization based on locally adaptive regression kernels algorithm
Kangjian He, Dongming Zhou, Rencan Nie, Xin Jin, Quan Wang · 2016
With the development of computer vision technology, it is possible to use different algorithms to detect and locate specific targets. Locally adaptive regression kernels (LARK), which is a method for rapid detection and localization of specific targets, is a kind of image analysis method without training, and can be used to search for specific target images quickly. It operates using a single example of an object as a query image to find similar matching in a target image. Firstly, features are preliminarily extra-cted from the query image using the local steering kernels as descriptors, and then salient features are further extracted using locality preserving projection (LPP). And then the comparison between the target image and the query image is done using the cosine similarity criterion to find all possible similar objects. Finally, the non-maxima suppression method is used to preserve the most similar objects, so as to achieve the detection and localization of the specific target. In this paper, we choose the different images with different angles and different scaling to carry out the experiments. The experimental results show that this algorithm is a good method for object detection and localization.