Foam infrared image segmentation combining NSST saliency detection and graph cuts

In order to accurately extract the collapse of the surface of the flotation foam, newly synthesized bubbles, and reduce the impact of noise and light, a foam infrared image segmentation method combining non-downsampling Shearlet Transform (NSST) domain saliency detection and graph cutting is proposed. NSST multi-scale decomposition of the foam infrared image, saliency detection of low-frequency subband images using GBVS algorithm, saliency value calculation based on Markov chain feature differences, noise coefficient removal and edge, weak edge coefficients are non-linearly enhanced. NSST reconstruction of the processed multi-scale high-frequency subband and low-frequency subband images is performed. The significance constraint term is constructed based on the saliency detection results of the low-frequency subband image, and the bubble brightness constraint term is constructed using a Gaussian fitting function. Then, the map is constructed Graph (Cut Cuts) energy function, and finally the maximum flow/minimum cut algorithm is used to segment the target area. The experimental results show that this method is less affected by light, and to some extent, it solves the problems of over-segmentation and under-segmentation. The detection accuracy rate of normal flotation under-flotation, and over-flotation are 91.8% 87.1%, 88.9%, respectively. The segmentation accuracy is significantly improved compared with the existing methods. It can effectively extract collapsed or newly synthesized bubbles, and shows good noise immunity. It has good robustness under different working conditions.