Local feature extraction method based on deep learning

Abstract

The invention discloses a local feature extraction method based on deep learning, and the method comprises the following steps: carrying out the network training, training a pre-constructed network on an image data set MS-COCO, segmenting the data set into a training set and a verification set which respectively comprise 82783 and 40504 images, carrying out the image matching, in an experiment, using a standard local feature pipeline to evaluate the performance of the local feature extraction method, wherein the standard local feature pipeline is used for extracting and matching features from any given pair of images in an experiment, then carrying out repeated score (Repeatability) calculation, then carrying out matching score (M-Score) calculation, and finally performing homography estimation effect evaluation. The detection step is postponed to description, so that more stable key points are obtained, compared with a traditional non-machine learning mode, the method has a more flexible feature searching process, and feature extraction precision is improved while a large number of key points are obtained.

Description

technical field [0001] The present invention relates to the technical field of local feature extraction framework of deep learning, in particular to a local feature extraction method based on deep learning. Background technique [0002] In many fields of computer vision, learning-based methods have emerged and began to outperform traditional methods. Intuitively, the feature extraction process only requires a network composed of several convolutional layers, which can simulate traditional detectors and detectors by learning appropriate parameters. Descriptor behavior. Some existing learning-based methods focus on training detectors or descriptors individually, while others successfully build end-to-end feature detection and description pipelines. For the former, when these individually optimized detectors Or the performance gains of these individual components may disappear when the descriptor is integrated into a complete pipeline, for which case it would be preferable to j...

AI Technical Summary

Problems solved by technology

[0003] However, it is challenging to achieve two different optimization goals by training a network, because the optimization goal of the detector is repeatability, while the optimization goal of the descriptor is differentiability, and the unification and joint combination of the two are very challenging. Without a good solution, existing technologies cannot balance these two optimization goals well

Images