复杂场景下全景视觉图像局部特征点检测仿真

Simulation of Local Feature Spot Detection of Panoramic Visual Images in Complex Scenes

为了在获取全景视觉图像特征点的同时有效保持图像的边缘信息,提出一种复杂场景下全景视觉图像局部特征点检测方法。采用经验模态分解方法分解全景视觉图像,在信息熵差值的基础上确定含噪的IMF分量,引入CLEAN算法抑制IMF分量中存在的噪声,实现全景视觉图像去噪处理;对全景视觉图像展开小波变换,消除图像特征图中存在的冗余信息,避免复杂场景对特征点检测产生影响,融合全景视觉图像的局部特征显著区域,获得显著图,完成局部特征点提取;根据特征点的独特性、均匀性、独特性、覆盖度挑选最优局部特征点,完成全景视觉图像局部特征点检测。实验结果表明,所提方法的去噪平滑性高、边缘保持能力强,特征点检测准确率高、效率高。

In order to obtain the feature points of panoramic images while effectively maintaining the edge information of images,this article presented a method of detecting local feature points of panoramic images in complex scenes.Firstly,the method of empirical mode decomposition was adopted to decompose a panoramic visual image,and then the noisy IMF component was determined based on the difference between information entropies.The CLEAN algorithm was introduced to suppress the noise in the IMF component,and thus to realize the denoising of the panoramic visual image.Secondly,wavelet transform was applied to the panoramic image for eliminating redundant information in the feature map,thus avoiding the impact of complex scenes on feature point detection.Moreover,salient areas of the local feature of the panoramic vision image were fused to obtain a salient image,thus extracting local feature points.According to the uniqueness,uniformity,and coverage of feature points,the optimal local feature points were selected to complete the detection of local feature points of the panoramic visual image.Experimental results show that the proposed method has high denoising smoothness,strong edge preservation,as well as high accuracy and efficiency of feature point detection.