Image segmentation quality evaluation method based on dM

Abstract

The invention relates to an unsupervised image segmentation quality evaluation method based on dM, and the method comprises the steps: carrying out the bilateral filtering and two-dimensional Gabor filtering of an original image, so as to extract a spectrum feature vector and a spatial feature vector; carrying out the transformation of the obtained spatial features, extracting three principal components, and carrying out the superposition with the spectrum features too form a spectrum-spatial feature set; taking the spectrum-spatial feature set as a base map, and calculating the layered spatial heterogeneity of the segmentation region at each dimension; obtaining the intensity value of the layered spatial heterogeneity, and calculating the spatial autocorrelation of the segmentation results of all dimensions; calculating a mean value for representing the intensity value of the layered spatial heterogeneity of each dimension in the feature set and the mean value of Moran indexes, and constructing the space of two values; calculating the Mahalanobis distance between each point in the space and a point (1, 0), and finally carrying out the evaluation of the segmentation quality of theimage according to the Mahalanobis distance. The beneficial effects of the invention are that the method improves the quality evaluation precision and stability, and an evaluation result accords withthe visual features of the human being in a better way.

Description

technical field [0001] The present invention relates to the technical field of image segmentation, in particular, to a d-based M Image segmentation quality assessment method. Background technique [0002] Image segmentation is a key technology in the entire field of image analysis. In the field of remote sensing images, image segmentation is a prerequisite for object-oriented image analysis. The segmentation results can be used for subsequent tasks such as object classification, target recognition, and scene understanding. Therefore, Accurate, quantitative and effective evaluation of segmentation results is of great significance for object-oriented remote sensing image analysis. At the same time, segmentation quality evaluation can further compare the performance of segmentation algorithms and optimize segmentation parameters to obtain better segmentation results. [0003] At present, image segmentation quality evaluation methods can be divided into five types: subjective e...

AI Technical Summary

Problems solved by technology

[0004] The unsupervised segmentation evaluation method does not need to provide an ideal segmentation reference standard, but establishes a specific quality standard based on human cognition to evaluate the segmentation results. The essence of image segmentation is the merging of homogeneous regions and the separation of heterogeneous regions. Therefore, The existing non-supervised evaluation methods mostly calculate the local and overall statistical values ​​of the segmentation area by designing statistical indicators, and then combine these indicators to obtain an overall quality score to evaluate the segmentation results. The non-supervised segmentation evaluation method is quantitative and objective. and high-efficiency features, but the design of statistical indicators, reasonable indicator joint strategy and the use of spatial information need to be further optimized

Images