Fractional ACM Image Segmentation Algorithms And Their Application To Weed Images

Image segmentation aims at partition the image into several homogeneous regions andextraction interesting target. Image segmentation is the premise of image understanding andimage recognition. Active contour model (ACM) based on partial differential equation (PDE),which utilizes curve evolution theory and level set method, takes advantage of the low-levelimage data information as well as high-level object prior knowledge with wide adaptability andscalability for natural scenes image. Fractional calculus is the expansion of Integer calculus,research on this aspect application in image processing is still in the early stage, currentlyresulting in only some preliminary success. The fractional calculus theory was introduced intoactive contour model, and ACM based on fractional differential image segmentation algorithmhas been systematically studied and discussed.On the basis of investigation of latest theoretical development, the ACM was classified andreviewed systematically. Various definitions of fractional calculus and design of fractionaloperator mask were described. Combined fractional differential operator with active contourmodel, the improved several segmentation algorithm was proposed and implemented, whichachieved satisfactory results in weed image processing. The main work was as follows:(1) Active contour model based on fractional differential. Classical CV model using onlyused the uniform regional information, and could not extract all target contour which the curvedidn’t converge to the desired boundary for weak edges or holes in the original image. The GACmodel used gradient and curvature for edge detection, which only made use of local edgeinformation, and the model was difficult to deal with discrete edge and sensitive to noise. For thedrawbacks of CV and GAC, With the GACV model a fractional order ACM was proposed andachieved stable object segmentation.(2) Fractional differential active contour models based on fractal dimension. Fractaldimension and multi-fractal dimension are important features of edge and area in the image,which has been applied to image segmentation, pattern recognition. Aiming at Fractional GACVmodel, the multi-fractal spectrum-H o&&lderindex was used to extract image edge features,Together with fractional order regularization term and region detection term, a fractional orderGACV image segmentation model with fractal dimension was realized for efficiently locating targets contours and achieving target detection.(3) Fractional differential active contour models coupled with noise removal. A fractionalACM image segmentation model was proposed using TV model and fractional order CV modelfor noise image. To overcome the non-convex of the energy functional in fractional TV andfractional CV models, a global minimum convex model was established with some constraintsand the existence of global minimum of the model was proved. Finally, the energy functional wassolved fast with split Bregman iterative algorithm. Experiments showed that this convexvariational model outperformed than the traditional level set algorithm, which could quickly andefficiently extract desired target without being sensitive to initial outline.(4) Application of Fractional ACM in weed image processing. Weed image recognition is oneof the key technologies of weeding robot, and weed image segmentation is the premise of weedrecognition. Under complex soil background in corn seedling field, the color corn images weresegmented by fractional ACM after feature extraction using linear manifold learning methods.Comprehensive comparison of various feature extraction algorithms, and image segmentationresults showed that the proposed color image segmentation algorithm coupled with maximummargin criterion (MMC) and fractional order ACM could obtain better results than traditionalcolor index and Otsu algorithm.