模拟视觉机制的隐蔽目标识别算法研究

Research on hidden-target recognition algorithm simulating visual mechanism

针对从复杂图像背景中识别隐蔽目标的问题。模拟动物视觉高效的图像信息处理机制,提出一种仅用少样本快速准确地从复杂图像背景中识别出隐蔽目标的算法。首先采用拓扑ICA方法从自然图像中学习初级视皮层中复杂细胞感受野;然后用含有隐蔽目标的图像及其待配准图像分别对神经元进行刺激,筛选出对刺激有强烈响应的神经元对应内容,并通过对比找出隐蔽目标对应特征,以此对隐蔽目标进行识别。用多幅自然图像进行实验,结果表明该算法耗时少、识别率高,达到了预期的效果。

Aiming at the problem of identifying hidden-target from complex image background, simulating the high- efficiency image information processing mechanism of animal vision, an algorithm has been put forward, which can use a little sample to identify hidden-target from complex image background quickly and accurately. Firstly, the re- ceptive field of primary visual cortex complex cell is studied from nature image adopting Topographic Independent Component Analysis method. Then the neurons are excited respectively by the images that contain the hidden-target and the image to be registered. The corresponding contents of the neurons that strongly response to the stimulations are screened out, the corresponding characteristics of hidden-target are found out through comparison analysis; and the hidden-target is recognized using the above method. Experiment results for natural images show the proposed al- gorithm consumes less time, has high recognition rate, and achieves the expected effects.