A dual-mode deep learning descriptor construction method based on graphic primitives

Abstract

The method is suitable for the technical field of image registration. The invention provides a dual-mode deep learning descriptor construction method based on graphic primitives. According to the method, attribute categories of patch images are learned by labeling samples, geometric features of the patch images are learned by utilizing graphic primitives, and the attribute categories and the geometric features are fused to obtain feature vectors of local patch images, namely descriptors based on the graphic primitives. Registration between the patches is completed by describing similarity of the sub-vectors. According to the dual-mode deep learning descriptor construction method based on the graphic primitives, which is provided by the invention, a descriptor classification method calculated by a GPU (Graphics Processing Unit) is explored aiming at the defect that a classic image registration method is relatively large in CPU (Central Processing Unit) calculation amount. A descriptor training set is mainly established, a multi-mode convolutional network is constructed, categories and geometric modes are trained on a GPU, and classification and registration of local patch images areachieved. The classification description method of the descriptors and the implementation problem on the GPU are solved.

Description

technical field [0001] The invention belongs to the technical field of image registration, and provides a method for constructing a dual-mode deep learning descriptor based on image primitives. Background technique [0002] Feature matching is the mainstream method of image registration. The classic image registration method uses local feature descriptors, which take the local area of ​​the image centered on the key point as the object, and describe the features of the area according to the grayscale information of its internal pixels. , and a feature vector expressing the local information around the key points of the image is obtained. However, classical descriptors are computationally intensive and difficult to apply to real-time systems, and are not suitable for mobile devices. Contents of the invention [0003] An embodiment of the present invention provides a method for constructing a dual-mode deep learning descriptor based on an image primitive, which aims to solv...

AI Technical Summary

Problems solved by technology

[0003] The embodiment of the present invention provides a method for constructing a dual-mode deep learning descriptor based on image primitives, which aims to solve the problem that the classic descriptor has a large amount of calculation and is difficult to apply to a real-time system

Images