In allusion to the problem of large computational complexity for the multi-scale Retinex algorithm in process of image en-hancement,the RBF neural network is proposed as a high contrast image enhancement algorithm, in which the feature vectors of 3 × 3 neighborhood pixels and the eigenvectors corresponding to the target image is obtained from the training data set and the center vector and the expansion constant of the network hidden layer is determined by the clustering algorithm and thus the optimal solution is acquired by the gradient descent method to make network converge quickly. The RBF neural network has been employed to establish the non-linear mapping relationship between the high contrast image and the enhancement algorithm and the image can be processed quickly,even in real time,according to the network parameters. The experimental results show that it has not only improved the edge and detail of the image but also promoted better sharpness of the image than traditional algorithms based on Retinex theory. Therefore,it is effective,which indi-cates better perspective in enhancement of high contrast image.%针对多尺度Retinex算法在图像增强过程中存在的算法运算量大的问题,提出了将RBF神经网络作为高反差图像增强算法.该算法从训练数据集中获取以3×3为邻域像素的特征向量以及目标图像对应的特征向量,通过聚类算法来确定网络隐含层的中心向量和扩展常数,采用梯度下降法使网络快速收敛得到最优解.利用RBF神经网络建立高反差图像与增强算法之间的非线性映射关系,根据神经网络参数进行快速图像处理,从而实现图像实时处理.仿真实验结果表明,与传统的基于Retinex理论算法相比,基于神经网络的高反差图像增强算法,不仅能够改善图像边缘以及细节,而且图像的清晰度也十分明显.因此,所提出的算法是一种有效的图像增强算法,在高反差图像增强中具有较好的应用前景.
展开▼
