| In the applications of mining engineering,quarrying and other related industrial activities,the traditional quality detection methods of rock particles mainly depend on the subjective vision of human,which will result in low efficiency and waste of resources.It’s the focus of current research area to improve the measurement accuracy of the size and shape of rock particles with machine vision and image processing technology.And the rock particle images are difficult to deal with in the field of image processing,therefore,it has a great practical value in the real-time detection of the mining engineering and quarrying to identify the rock particles from images.This paper introduces the principle of NSCT and graph theory in details.The research mainly focus on using the knowledge of NSCT and graph to realize the image segmentation of rock particle images with different qualities,which can be divided into image pre-processing and segmentation.The contents are as follows:1.Rock particle classification and image pre-processing.There are differences in the characteristics of different types of rock particles,and the difficulty to recognize them from the images is not the same,therefore,it is necessary to classify them for subsequent judgment.The environment of rock engineering is relatively bad,the quality of rock particle images is generally affected by noises,which may impact the follow-up work.This paper uses NSCT to decompose the image and only take the energy of the first two scales to eliminate the impact of noise by comparing the common de-noising methods.2.Improved Watershed pre-segmentation.The processing of a large scale rock particle image is quite time-consuming,in order to improve the processing speed,the dissertation uses a Watershed algorithm which combines with the knowledge of morphology to segment the image roughly by considering the advantage of less time-consuming and forming small areas,and applies the color information to eliminate the border lines generated by the Watershed algorithm.3.Improved Normalized Cut segmentation algorithm.The traditional Normalized Cut costs a lot of time,therefoe,the Watershed algorithm is applied to improved the Normalized Cut to reduce the cost of time,the areas are viewed as nodes to form a graph,and then,the weight function of the Normalized Cut is determined by the color information of the regions and the spatial relationship between the regions,which can help to achieve a better segmentation results of different types of rock particle images,this method is able to deal well with the images with little noise.4.Noisy image segmentation based on NSCT and graph theory.In the practical application of engineering,we hope to have a special method to deal with the low quality images,since the improved Normalized Cut method is not able to deal well with the image polluted by the noises severely,therefore,we propose a new algorithm,in which,we firstly use an adaptive threshold function based on NSCT domain for image de-noising,and then,the weight function of the Normalized Cut is improved with the gradients of pixels in order to achieve the good segmentation results of the polluted rock particle images. |
