基于眼动跟踪的视频视觉感兴趣区域提取仿真

Simulation of video visual region of interest extraction based on eye tracking

海量视频中具有显著特征的区域为感兴趣区域。由于视频图像中普遍存在较多单目标特征,轮廓边缘关联性较强,干扰视觉区域的提取。于是提出基于眼动跟踪的视频视觉感兴趣区域提取方法。提前转换原始视频颜色空间,降低色彩影响程度,检测原始视频边缘,利用形态学闭运算填补视频边缘断裂位置并去噪,通过旋转校正处理得到简单完整的视频图像,利用眼动跟踪处理技术提取纹理特征以及亮度显著图,实现视频视觉感兴趣区域提取。实验结果表明,研究方法能够精准提取出视频视觉感兴趣区域,峰值信噪比最高可达1.5dB,验证了该方法的应用精度和图像质量增益均较高。

Regions with significant features in videos are the region of interest.Because there are many single target features in video images,the contour edge has a strong correlation,interfering with the extraction of the visual region.Therefore,a method of extracting visual regions of interest in video was proposed.Firstly,we converted the color space of the original video in advance,thus reducing the influence of color.Secondly,we detected the edge of the original video,and used morphological close operation to fll the break of the video edge and thus remove noise.Thirdly,we used the rotation correction to get a simple and complete video image.Meanwhile,we used eye tracking technology to extract texture features and brightness saliency maps.Finally,we extracted the visual regions of interest.Experimental results show that the proposed method can accurately extract the visual region of interest,and the peak signalto-noise ratio can reach up to 1.5dB,which verifies that the application accuracy and image quality gain of the method are high.