【摘要】 针对复杂背景的分形尺度不变性,提出了一种扩展目标分割新算法。首先将大、小两种尺度下像素邻域内的灰度信号转化为一维,分别进行离散傅立叶变换(DFT),再利用频域的极大似然估计提取分形维数,选取小尺度的分维数图表征各边缘的精确位置,而大尺度的分维数图表征抑制背景纹理后的目标主要轮廓,然后计算图像边缘的方向性函数,使大尺度边缘在小尺度边缘限定的范围内,选择幅度与方向接近的目标边缘进行连接,最后进行阈值分割,得到闭合的精确边缘结果。实验证明了算法的有效性和可靠性。更多还原

【Abstract】 Extended target varies little in multi-scale. It is proposed to transform intensity of pixels in two domains into one-dimension signals and have corresponding Discrete Fourier Transform. By maximum likelihood estimators in Fourier domain, we can get two fractal dimension images. The bigger-scale fractal parameters represent main contour of target after restraining noise and sharp edges of background. The smaller-scale ones would denote precise locations of edges. According to information of edge directions, the remaining edges with similar amplitude and direction would join together. At last, closed contour of target is gained. Abundant experiments prove that the scheme is efficient and credible.更多还原