基于LVQ神经网络的水果图像分割研究

由于传统边沿检测算子在水果颜色多样、亮度不均匀时,难以分割得到完整、无噪声的二值图像且依赖优化的阈值,本研究提出了一种基于LVQ神经网络的水果图像分割方案。首先将彩色图像转变为灰度图像;然后对Canny算子获得的边沿图像随机选取一些像素作为网络的学习监督信号,仅以灰度图像中相同位置像素3×3邻域的Kirsch算子梯度值作为输入,训练权值;最后重新将原灰度图像的Kirsch算子梯度值输入到训练好的网络中,获得封闭的边沿并填充得到二值图像。考察了14幅像素为640×480的水果图像,结果表明：网络在很宽广的阈值范围内（0.001 ~ 0.99）分割得到完整、一致的二值图像;面积误差最小为0.9%,最大为8.83%,不依赖于优化的阈值,不需要对原始图像滤波预处理。与没有阈值及滤波的算法相比,本方案的误差和时间复杂度均更低;与设置了阈值和/或滤波的算法相比,本方案与之相当,甚至效果更优。

FRUIT IMAGE SEGMENTATION BASED ON LVQ NEURAL NETWORK

To solve the problem that the traditional edge detection operator is difficult to obtain a complete binary image when the color of fruit is diverse, and the brightness is uneven, and it depends on the optimized threshold, a new method of fruit image segmentation based on simplified LVQ neural network is proposed in this paper. Firstly, the color image is transformed into gray image. Then some points from the edge image obtained by Canny operator are randomly selected as the learning supervision signal, and only the gradient value of Kirsch operator in the 3 × 3 neighborhood of pixels at the same position in the gray image is taken as the network input, and the weights are trained; Finally, the gradient value of Kirsch operator of the whole gray image is re-input into the trained network to obtain a closed edge, which is filled into a binary image as the segmentation result. Fourteen fruit images with 640 × 480 pixels are selected for investigation. The results show that complete, consistent and noiseless binary images are segmented within a wide threshold range (0.001- 0.99); The minimum area error is 0.9% with the maximum 8.83%. They are not depended on the optimized threshold and pre-filtering. Compared with algorithms without thresholds and filtering, this scheme has lower errors and time complexity; compared with algorithms that have set thresholds and/or filters, this scheme is comparable or even better.