【摘要】 传统视频图像异常区域检测与推荐的过程,忽略了对图像的非线性映射处理,导致图像噪声较大,异常目标区域推荐精度偏低,提出网络数字视频图像异常目标检测区域推荐方法。将网络数字视频图像设定为网络的输入、输出图像,利用隐含层组建由含噪图像到去噪图像的非线性映射,通过卷积子网与反卷积子网组建对称的网络结构,结合修正线性单元激活函数完成图像去噪。提取图像图像低频子带图像及特征向量,通过支持向量机实现对特征向量的分类,完成网络数字视频图像异常目标检测区域推荐。实验结果表明,优化后方法具有更高的图像异常目标分类、检测精度,且时间开销更小。更多还原

【Abstract】 In the traditional detection and recommendation for abnormal region in video images, the non-linear mapping of image was ignored, leading to high image noise and low recommendation accuracy. Therefore, a recommendation method for abnormal target detection region in network digital video image was presented. The network digital video image was set as the input and output images. The hidden layer was used to construct the non-linear mapping from noisy image to denoised image. The symmetrical network structure was constructed through the convolutional subnet and the deconvolution subnet. Combined with the activation function of modified linear unit, the noise reduction of image was completed. The low-frequency sub-band images and feature vectors were extracted. Finally, the feature vectors were classified by support vector machine, and the abnormal target detection region of network digital video image was recommended. Simulation results show that the optimized method has higher classification accuracy and lower time cost.更多还原