An Image Merging and Segmentation Method Based on Region Adaptive Spectral Angle Threshold

Abstract

The invention provides an image merging and segmentation method based on the area self-adaptive spectral angle threshold, which can avoid over-segmentation and under-segmentation and realize effective image segmentation. The invention is a target-oriented image segmentation algorithm, which uses the average spectral angle between regions with intuitive physical meaning as the basis for judging whether to merge or not and the merge sequence. The spectral angle is used as an important and unique input parameter for global gradient calculation, over-segmentation generation, initial judgment of global object similarity, and generation of region-adaptive spectral angle threshold, which fully considers the spectrum of color, multi-spectral and even hyperspectral images in the field of remote sensing. Correlation. Using the spatial statistics of the merged region, the region to be merged and the boundaries of the two—regional homogeneity, the spectral angle threshold suitable for each region is adaptively generated. It not only fully considers the spatial correlation of the image, but also avoids the uncertainty of the spatial relationship caused by the irregular geometric shape of the space segmentation.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to an image merging and segmentation method based on a region adaptive spectral angle threshold. Background technique [0002] Traditional image segmentation methods are mainly divided into region-based methods and boundary-based methods. The main difference between region-based and boundary-based is to use spectral similarity or difference to generate different adjacent blocks. Since the traditional method of interpreting and dividing images takes pixels as the research unit, it is always difficult to avoid over-segmentation or under-segmentation when dealing with target-oriented segmentation problems. In order to overcome the above shortcomings of traditional methods, scientists proposed a processing framework that first generates over-segmented images and then merges similar blocks in them. [0003] Common methods for generating over-segmentation are based on support v...

AI Technical Summary

Problems solved by technology

However, the above research still uses one or more discrete scales to segment the image in essence, and does not solve the problem of discontinuous segmentation scales in object-oriented image segmentation, nor can it fundamentally avoid the problems of over-segmentation and under-segmentation

In 2017, Yang, Jian et al. proposed an algorithm framework based on the spectral angle threshold to segment first and then merge. However, this method only considers the properties of each region itself in a single merge, and ignores the relationship between the current research region and adjacent regions. However, the connection between them will still lead to the emergence of over-segmentation and under-segmentation

Images