An object detection method based on discriminative semantic component learning

Abstract

The present invention provides an object detection method based on discriminative semantic component learning. The discriminative semantic component clustering and purification based on sparse representation proposed by the present invention obtains the final semantic component set, and the detection result is obtained by using the object confidence spectrum of the learned semantic component detector. Compared with the existing component-based object detection methods that require strong supervision information during object model training and can only detect specific types of objects for training, the semantic component learning process of the present invention is fully automated, and in the entire component learning process, only The window labeling information of the object is required, and there is no need to provide more strongly supervised component information; the geometric similarity of different types of object component information is used, and based on the sharing of different types of object components, it can have geometric similarity for cross-category pairs The generalization performance of component-based object detection algorithms is improved.

Description

technical field [0001] The invention proposes a discriminative semantic component learning method to solve the object detection task in the image, which is a new technology in the field of image processing. Background technique [0002] Nowadays, with the continuous development and maturity of computer network, multimedia technology and digital media equipment, people's demand for digital images in work and daily life has also increased significantly. Among the massive digital images, in order to further analyze and process the image data, people often need to locate the object area of ​​interest in each image, and then the object detection technology was born. The object detection problem is to design an effective algorithm to identify and locate the object area of ​​interest in the input image data. In the face of the current massive image data information, the object detection method provides an effective way to analyze and understand the information content in the image...

AI Technical Summary

Problems solved by technology

However, the shortcomings of this type of method at present are that, in specific detection, only the currently trained object categories can be detected, and for new object categories, even if there is geometric similarity between objects, it cannot effectively detect It detects and identifies

Therefore, for this reason, the practical application ability of such methods is also limited.

In addition, general class object detection methods utilize low-level visual cues of image data, such as image-based segmentation and salient features, which overcome the problem that specific object detection methods can only complete single-class object detection, but the detection accuracy of this class of methods Not as good as class-specific object detection methods

Images