小波图像修补技术在小麦NIR图像处理中的应用

Application of Wavelet Image Inpainting in Wheat NIR Image Processing

在利用小波变换对小麦近红外图像处理的过程中存在小波系数缺损的问题,小波图像修补技术可以有效地恢复丢失小波系数。为此,针对小麦的近红外图像,引入了基于全变差(TV)的小波图像修补技术,分别对随机丢失5%和50%的小波系数的小麦种子进行近红外图像修补后,峰值信噪比PSNR分别由23.83dB和9.96dB提高到36.81dB和33.20dB。修补后的近红外图像中,小麦的轮廓及腹沟部分基本修补到接近原始图像,在种子的果毛、胚等包含纹理细节的部分修补效果不够理想。实验表明,基于全变差的小波图像修补技术可以恢复小麦近红外图像在处理过程中丢失的大部分系数,从而使得图像保存的信息更加完整。

During the processing of NIR images of wheat by wavelet transformation, the image quality is often influenced because of the losing of wavelet coefficients. Wavelet image inpainting could effectively reconstruct the lost coefficients. In this study, the total variation （TV） based wavelet image inpainting is applied to process the NIR images of wheat seeds. The wavelet coefficients were randomly lost 5% and 50%, respectively. After inpainting, the Peak Signal to Noise Ratio （PSNR） of the corresponding reconstructed images were improved from 23.83dB and 9.96dB to 36.81dB and 33.20dB, respectively. The profile and ventral groove of reconstructed wheat images were close to original images. The inpainting effect of the detail part such as seed hair and embryo were not perfect. The results showed that the total variation （TV） based wavelet image inpainting could get back most lost wavelet coefficients during the process of NIR images of wheat, thus make the preserved images have more information.