视觉假体中动态图像识别研究

为了确定低分辨率动态图像识别的最小信息需求，通过灰度化、二值化、边缘提取和匹配不同的仿真光幻视模板等处理策略将绘画过程视频处理为五种分辨率（24×24，32×32，48×48，64×64和128×128）的像素化动态视频，对其进行视频复杂度分析，并使用简化的Itti算法提取特征点来分类有效信息，记录并统计分析不同分辨率下的被试者识别时间以及识别准确率。实验结果表明，随着不同分辨率下的视频复杂度的升高，识别时间逐渐减少，识别准确率不断升高；动态视频的像素化分辨率越高，识别所需特征信息越少，当视频像素化分辨率达到64×64或128×128时，被试者只需少量特征信息即可完成识别。

Study on dynamic image recognition in visual prosthesis

In order to determine the minimum information requirements of the recognition of dynamic images with low resolution, the strategies such as grayscale, binarization, edge extraction and matching simulated phosphene different template are used to process the painting videos and form five kinds of resolution（24×24, 32×32, 48×48, 64×64 and 128×128） dynamic video material library. The complexity analysis of videos is performed and the feature points are extracted and classified by using the simplified Itti algorithm. The experimental results of different resolution identification time and recognition accuracy are recorded and analyzed. It shows that with the increase of the complexity of different resolution video, recognition time is gradually reduced and the recognition accuracy is increased. Less information is required to identify the feature with increase of pixel resolution. Especially, when the video pixel resolution reaches 64×64 or 128×128, subjects can complete the identification task by a small amount of characteristic information.