Research On Agricultural Image Denoising Algorithm Based On Hybrid Wavelet Transform

There will involve numerous image processing operations in agricultural informatization, intelligence, auto grading and inspection, and machine vision fields. Image denoise is one of the most important image pre-processes which plays a significant role in agricultural information and intelligence. The performance of denoising algorithm will directly affect the level of automated inspection. Wavelet transform has been widely used in image denoise, and has achieved significant results for its characteristics of time-frequency composite, multi-resolution analysis and uncorrelation. However, wavelet denoise is approximating optimal rather than optimal in signal space. Moreover, the current denoising methods used in agricultural image such as Median filter, Gaussian filter and Average filter have poor visual effects and low peak signal to noise ratio (PSNR). This would go against the images'high level understanding and analysis. With the constant improvement of agricultural informatization, it has been a necessary requirement to research a new denoising algorithm which could perform very well in agricultural images.This paper is supported by the National Natural Science Foundation of China (30971690), taking the research works as follows:(1) Inspired by the cross breeding and learned the idea of generic algorithm, this paper proposed an image denoising method based on hybrid wavelet transform, which was applied in agricultural images. This algorithm takes the image denoised by adaptive wavelet threshold as male parent and image denoised by wiener filter as female parent. The male and female parents are coded to be two initial populations. Hybridize and mutate the individuals which are selected by selection rate, and expect their offspring to perform much better characteristics. The offspring which meet the best standards will be decoded to be the optimal solution, and restored to be image. Then, the offspring image will be regarded as the best result of the algorithm. The agricultural images of red jujube, wheat and apple were used to verify the algorithm. Experiment results show that hybrid wavelet transform algorithm could suppress the noise much better than traditional denoise methods. The denoised images'PSNR can achieve to 71.8875 for red jujube, 75.0014 for wheat image and 70.5784 for apple image (PSNR of noise image are 59.6517, 61.5131 and 59.6673); Applied the algorithm to Lena and Camera images, and the processed image also have a good vision effect and high PSNR.(2) The effects of wavelet denoising with different wavelet base function were studied and analyzed. Because the wavelet base function could affect the sparse level of wavelet coefficients when the image was transformed by wavelet transformation, the denoising effect would also be affected. Therefore, the wavelet base function which could sparse the wavelet coefficients better should be selected for adaptive wavelet threshold denoise.(3)The hybrid wavelet transform uses male parent to hybridize the female parent. So, different male parent will affect the denoising results seriously. The research on selecting male parent has been done for the algorithm to select an optimal male parent. Wiener filter, Gaussian filter, Median filter and Average filter has been used to verify the algorithm. Experiment results show that Wiener filter could achieve better results as male parent than other three filters, for their PSNR were lower than noise image. Because that the processed images were seriously damaged with white spots and vertical lines.(4) Build and implement the image denoising analysis system with Visual C++ 6.0 and computer vision library OpenCV. The system realized image's adding noise, traditional denoise, hybrid wavelet denoise, analysis of denoise methods and wavelet transform. Image adding noise model could add Gaussian noise to image with certain mean and variance input by user. Hybrid wavelet denoise model denoised the image with hybrid wavelet denoising methods by the input image of male and female parents. Analysis of denoise methods model combined a variety of currently used denoising methods, and showed the process results and PSNR on the same control panel. This will facilitate the researchers to analyze and summarize the different denoising methods.