A foam infrared image segmentation method based on nsst saliency detection and graph cut

Abstract

The invention relates to a foam infrared image segmentation method based on NSST saliency detection and graph cutting. First, the foam infrared image is decomposed by NSST to obtain a low-frequency sub-band image and a multi-scale high-frequency sub-band; The sub-band image is saliency detected, and the saliency value and visual saliency map are obtained; then, the threshold and scale correlation coefficient are calculated for each high-frequency direction sub-band coefficient, and the noise coefficient and nonlinear enhancement edge and weak edge coefficient are removed. NSST reconstructs the image, and combines visual saliency to perform graph cuts to obtain segmentation results. The method has strong anti-interference ability and high segmentation accuracy.

Description

technical field [0001] The invention relates to the technical field of flotation, in particular to a foam infrared image segmentation method based on NSST saliency detection and graph cutting. Background technique [0002] Flotation is the collision and adhesion of minerals in the flotation machine and the micro-bubbles in the air. Using the hydrophilic and hydrophobic properties of the surface of minerals and impurities, the mineral particles with high floatability float to the foam layer of the liquid surface with the bubbles, so as to remove the target minerals. A beneficiation method for sorting ores with complex material composition. Studies have shown that the generation and collapse of bubbles during flotation can effectively reflect the mineral content, and image segmentation is the key to flotation image processing to analyze and detect newly generated and collapsed bubbles, which are widely used in computer vision and industrial production. [0003] The collapse r...

AI Technical Summary

Problems solved by technology

However, the foam infrared image has the problem of low resolution and contrast, and is easily disturbed by noise, and the high temperature area at the edge of the bubble will affect the segmentation results. At present, the segmentation methods of infrared images mainly include: the region growing segmentation algorithm first performs Gaussian filtering to remove noise, and then Using the region growing algorithm, the segmentation accuracy and speed are improved, but it will cause holes and blurred edges in the segmented area; the improved watershed segmentation method uses area reconstruction transformation to solve the problem of difficult extraction of watershed identification points, but there is a problem of blurred boundaries And the segmentation parameters need to change with the actual situation; the threshold segmentation algorithm extends Otsu threshold segmentation to Otsu multi-threshold segmentation, and introduces the group search random optimization algorithm into multi-threshold segmentation, which improves the threshold search speed and segmentation accuracy, but is not suitable for noise processing and Processing of infrared photovoltaic images with complex backgrounds; K-clustering algorithm segmentation first enhances infrared images based on morphology, and then performs image segmentation by modifying the distance formula used in k-means clustering. The segmented image target outline is clear, But the integrity of the target area is low

Images