Gradient and color characteristics-based automatic straight line matching method in digital image

Abstract

The invention relates to an automatic straight line matching method in a digital image. The method comprises the following steps of: acquiring an image and inputting the image into a computer; extracting straight line segments in the image; determining a linear neighborhood and dividing the linear neighborhood into sub-regions; respectively performing inner product operation and outer product operation on the Gaussian gradient of each point and the average Gaussian gradient of the sub-region in which the points are positioned to acquire a gradient characteristic; performing normalization on the color of each point and the average color of the sub-region in which the points are positioned to acquire a color characteristic; computing gradient mean descriptor vectors and gradient standard deviation descriptor vectors of the straight line segments based on the gradient characteristic; computing color mean descriptor vectors and color standard deviation descriptor vectors of the straight line segments based on the color characteristic; computing euclidean distances between every two descriptor vectors of the straight line segments, summating after assigning different weight coefficients to the euclidean distances, and acquiring the similarity between the straight line segments; and matching the straight line segments and outputting a result. The method provided by the invention utilizes two image characteristics of the gradient and the color at the same time without needing any prior condition, can be applied to different types of images, and has more extensive adaptability.

Description

technical field [0001] The invention relates to the field of automatic matching of image features in computer vision, in particular to a method for automatic matching of straight lines in digital images. Background technique [0002] Feature matching technology has important applications in the fields of image retrieval, object recognition, video tracking and augmented reality. In recent years, marked by the proposal of the scale-invariant feature transform (SIFT) technology, the automatic matching technology of image feature points has made great progress. However, because the length of a straight line is not easy to be described uniformly, and the texture near the straight line is generally not rich, the automatic matching technology of straight lines is rarely proposed. [0003] At present, the main line matching methods are mainly as follows. The first is the method based on the color histogram. This method only uses the color information near the line to construct the ...

AI Technical Summary

Problems solved by technology

However, since the length of a straight line is not easy to describe uniformly, and the texture near the straight line is generally not rich, the automatic matching technology of straight lines is rarely proposed.

[0003] At present, the main line matching methods are mainly as follows. The first is the method based on color histogram. This method only uses the color information near the line to construct the histogram and perform matching. It cannot be applied to gray areas with indistinct color characteristics degree images, remote sensing images, etc.; the second is the mean standard deviation line descriptor (MSLD) method. When performing line matching, this method only uses the gradient information in the image to construct a line descriptor for matching, and does not use color Information, the matching performance of the image type with significant color features is not good; the third is the line clustering (Line Signatures) method, which uses the ratio of the angle and length of the line to match the line, and relies on the accurate extraction of the end points of the line, but For image types with rich textures, it is difficult for general line detection algorithms to obtain the exact position of the endpoints. This method is only suitable for image types with simple scenes; the fourth is a method based on feature point matching, which requires first Matching, and then use the position constraints provided by the feature points to perform line matching, but for image types with simple textures, the feature point matching algorithm can obtain fewer matching pairs, resulting in poor final performance of the algorithm

[0004] In specific applications such as image recognition, it is generally impossible to determine the image type in advance, and some task scenarios need to process multiple types of images at the same time, which requires that the line matching algorithm can be applied to various image types

However, the existing line matching methods are only limited to specific image types and are not universal.

Images