A rectangle detection and extraction method based on contour matching optimization

Abstract

The invention relates to a rectangle detection and extraction method based on contour matching optimization, which is characterized in that it comprises the following steps: Step 1: Detect and extract the edge contour of the original image to obtain the edge image; Step 2: Calculate each non-edge pixel in the edge image The distance D(p) from the nearest edge pixel, and then perform normalized distance transformation to obtain the normalized distance S(p), where p is a non-edge pixel; Step 3: use a regular grid to cover the edge image, and Assign initial values ​​to the rectangles in the regular grid; step 4: calculate the comprehensive contour matching degree of each rectangle in the regular grid and the rectangles in the rectangle's neighborhood set, and select the rectangle with the largest contour matching degree as the initial rectangle extracted; step 5: Remove repeated rectangles and non-rectangles to obtain the final extraction rectangle of the edge image. The method of the invention regards the rectangle as a whole, expresses it as a five-dimensional vector parameter, and realizes high-precision detection and extraction of the rectangle by optimizing the matching degree between the outline of the rectangle and the edge extracted from the image.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, in particular to a rectangle detection and extraction method based on contour matching optimization. Background technique [0002] The shapes of a large number of artificial objects appear as rectangular features, such as street signs and license plates in natural scene images, and cars, buildings, and salt fields in high-resolution remote sensing images. This rectangular feature is significantly different from the shape with fractal features in natural scene objects, and provides excellent features for automatic detection and extraction of these objects. Especially with the launch of a number of high-spatial-resolution remote sensing satellites such as my country's Ziyuan-3 and Gaofen-2, a large number of high-resolution remote sensing images that urgently need real-time processing have been obtained. However, the current object-oriented methods based on segmentation often enco...

AI Technical Summary

Problems solved by technology

However, this method is slow due to the pixel-by-pixel sliding window strategy used in the method

In addition, conditions such as edge bending will also cause the failure of straight line detection, which in turn will lead to missed detection of rectangles.

Other strategies that rely on the combination of straight line detection results (Kawatani Yamato, Kosaka Kiyoto, Wu Teng Hiroyasu. Image processing device and rectangle detection method: China, 201210439013.X[P].2012-11-06.), usually also You will encounter missed inspections like edge bending and breakage

Images