Brain magnetic resonance image super-voxel generating method based on prior knowledge

Abstract

The invention discloses a brain magnetic resonance image super-voxel generating method based on the prior knowledge. The method comprises the following steps: taking the weight of the space distance, the pixel intensity and the prior knowledge as the final distance measurement based on a K-means clustering algorithm, clustering the image pixel, and segmenting the brain MRI image into a series of uniform super-voxels capable of preferably fitting with the image edge. By integrating with the prior knowledges of different tissues of the brain, a new measure operator is designed, a robust super-voxel generating method is constructed to realize the super-voxel segmentation of the brain magnetic resonance image, and the influence on the segmentation result by the image noise can be reduced. Compared with the known super-voxel generating method, the method disclosed by the invention has high efficiency, higher boundary fitting degree, and is capable of preferably inhibiting the influence of the noise.

Description

technical field [0001] The present invention relates to the technical field of digital image processing, and relates to a processing method of brain magnetic resonance images, and more specifically, relates to a supervoxel generation method of brain magnetic resonance images based on prior knowledge. Background technique [0002] Superpixel (superpixel) or supervoxel (supervoxel) is an image preprocessing technology that has developed rapidly in recent years. Compared with the basic unit in traditional processing methods—pixels, superpixels are more conducive to the extraction of local features and the expression of structural information, and can greatly reduce the computational complexity of subsequent processing. It has been obtained in the field of computer vision, especially in image segmentation. widely used. In addition, superpixels have also been introduced into other computer vision research fields such as object tracking, human pose estimation, object recognition,...

AI Technical Summary

Problems solved by technology

[0004] However, for different types of images, current superpixel generation algorithms generally have mutual constraints between the number of superpixels, compactness, segmentation quality, and algorithm practicability.

At the same time, it is difficult to obtain better segmentation results for special targets.

Especially for the characteristics of MRI with large noise and weak boundaries, the above algorithm is not very suitable, especially when the image is seriously disturbed by noise, it cannot get a good superpixel generation effect

Images