基于Walsh基和二值离散余弦基单像素成像的对比分析

为提升单像素成像的测量效率,对Walsh基和二值离散余弦基单像素成像进行了对比研究。针对Walsh变换和抖动二值化后的离散余弦变换,提出了一种借鉴机器学习的排序方法,即通过分别统计80000幅图像的Walsh测量基和离散余弦测量基各自相应测量权重绝对值,按权重由大到小排序得到了两种具有先验知识的测量基优化排序。按两种测量基优化排序,研究了在相同压缩采样下的单像素成像。利用随机选取的500幅图像,定量分析了成像的峰值信噪比和结构相似度。研究发现,两类测量基在压缩采样为5%时,500幅图像平均峰值信噪比可达20dB,平均结构相似度达0.5以上;在采样率1%~10%的区间,二值化的离散余弦基重建结果占优;在采样率10%~20%的区间,两种测量基效果相当;在采样率大于20%时,Walsh基占优。该结论为单像素成像技术实现快速、高信噪比成像提供了一种方法,对推进单像素成像测量基的选取具有较好的参考价值。

Walsh Bases and Binary Discrete Cosine Bases Single-pixel Imaging Comparing

In order to improve the measurement efficiency of single-pixel imaging, a comparative research is performed on Walsh bases and binary discrete cosine bases single-pixel imaging. Aiming at the Walsh transform and the dithered discrete cosine transform, a sorting method based on machine learning is proposed. The weights of 80000 images are measured respectively by the Walsh bases and discrete cosine bases. The absolute valuesof weights are sorted from large to small, thus two kinds of measurement bases optimization rankings with prior knowledge are obtained. Single-pixel imaging under the same compression sampling are studied by these two kinds of measurement bases. Using 500 randomly selected images, the peak signal-to-noise ratio and structural similarity of the images are quantitatively analyzed. The research results shown that the average peak signal-to-noise ratio of 500 images is as high as 20dB and the average structural similarity is above 0.5 when the two types of measurement bases are compressive sampling at 5%. In the interval of 1%~10% sampling rate, the binary discrete cosine bases has the better results. When the sampling rate is 10%~20%, the two measurement bases have the same effect. When the sampling rate is greater than 20%, Walsh bases is dominant. This conclusion provides a method for single-pixel imaging technology to achieve fast and high signal-to-noise ratio imaging, and has a good reference value in advancing the selection of single-pixel imaging measurement bases.