Target Segmentation in Complex Environment Using Fractal Features

By analysis of the discrete fractal Brownian random field model, an intelligent segmentation algorithm is proposed to process targets in complex environment. Firstly, to smooth the rough background texture, four-direction gradients are extracted out for filter which would obviously reduce singular values with variable gray intensity distribution. Secondly, a new fractal parameter, named fractal modulation degree, is computed out to highlight immanent diversities of target and background. Then, passing through three-layer BP NN, multi-features are trained to obtain rational weight values and perform pattern recognition. Eventually, the contour of target is segmented out. Abundant experiments support the scheme´s satisfying validity and reliability