Research of Image Main Objects Detection Algorithm Based on Deep Learning

Abstract

Images have many objects in complex background. How to identify these objects and identify the main objects therein and understand the relationship between the main objects and other objects are the focus of this parper. There are many ways to object recognition, but most of them cannot mark the main objects of the image. In this paper, we use improved RCNN [1] network to detect and recognize multi-object in the image. Then we put forward the main objects scoring system to mark the image main objects. The experimental results show that the algorithm not only maintains the superiority of RCNN, but also detects the main objects of the image. We found that the image main objects were related to the size of candidate region and the rarity of objects.