基于二维Tsallis熵的改进PCNN图像分割

为了改善图像分割的性能,采用改进的脉冲耦合神经网络(PCNN)进行分割,通过对其内部活动项进行空不变的单阈值化分割,来达到对原图像空变阈值化分割效果.另外分割准则也作了修正,通过计算图像二维直方图的Tsallis熵,得到二维Tsallis熵,以此作为图像分割准则.最后,修正了动态门限项的下降速度,使得PCNN收敛更快.实验证明二维Tsallis熵准则优于最大Shannon熵准则与最小交叉熵准则,且改进的PCNN模型比传统PCNN模型收敛更快.

Image segmentation based on 2D Tsallis entropy with improved pulse coupled neural networks

In order to ameliorate traditional image segmentation,an improved pulse coupled neural network(PCNN) is introduced.The inner function item of the PCNN is looked as a new image,which is segmented with space-invariant threshold to achieve the effect that original image is segmented with space-variant threshold.Meanwhile,Tsallis entropy is combined with 2D histogram to engender 2D Tsallis entropy,which is used for guiding image segmentation.Finally,the decrease velocity of dynamic threshold is accelerated in order to make the PCNN converge faster.Experiments show that 2D Tsallis entropy performs better than maximum Shannon entropy and minimum cross entropy,and the improved PCNN can converge more quickly than conventional PCNN.