基于简化PCNN模型的结构光图像自动分割方法

Simplified PCNN-based automatic segmentation method for structured light images

针对CCD获取的结构光图像因大尺寸、光照不均匀,一般分割方法容易产生过分割或欠分割,提出了一种简化的脉冲耦合神经网络(PCNN)分割方法。将结构光图像进行分块,降低光照对分割质量的影响。每块子图像采用改进的PCNN模型自动进行分割。PCNN采用线性方式动态调整脉冲门限,以最小交叉熵确定其迭代次数,并利用邻域像素间的关系自动调整连接系数,减少人工干预。通过主客观评价指标对分割结果进行了比较,结果表明,提出的算法可以有效地分割出结构光图像中的条纹及点阵模式,目标边缘光滑、连贯和清晰,可以用于结构光图像的分割处理。

An image segmentation algorithm based on simplified pulse coupled neural network（PCNN） is proposed,which aims to overcome the over or under segmentation in some traditional segmentation algorithms for structured light images with large size and uneven illumination,and captured by CCD.Block processing for structured light images was used to reduce the influence of illumination on segmentation quality.Each block-image was automatically segmented by the improved PCNN.The PCNN dynamically adjusted pulse threshold using the linear mode.The minimum cross-entropy is used to determine the number of iterations.The relationship between neighborhood pixels was used to automatically adjust the connection coefficient,which could reduce manual intervention.Subjective and objective evaluation indexes of the segmentation were compared.Experimental results show that the proposed algorithm can effectively segment the stripes or dot patterns of structured light images.The edge of segmentation target is smooth,consistent and clear,which verifies that the PCNN can be used to segment structured light images.