Research On A Hybrid Wavelet Transform-based Red Jujube Image De-noising And Defect Detection

The red jujube is one of the specific fruits in our country, and its variety resources areabundant. Besides, the red jujube is one of the preponderant agricultural products in ShaanxiProvince. However, as the red jujube grows in natural environment, there is often some noiseon it. The noise can affect the accuracy of its defect detection and automatically grading. Inaddition, influenced by the weather and environment, the red jujube often has defects such asscab, flaw, and mould. These defects will have bad influence on the selling of red jujube.Therefore, in order to increase the processing level after the red jujubes are picked, at thesame time, to overcome the shortage of manually grading, we need to use machine vision,which has been widely applied. In machine vision, image denoising is a very importantpre-processing step. And detecting defect in red jujube exactly through image processing andanalysis technique is the key point in red jujube quality automatic detection.Based on this background, the research aim in this paper is to remove the noise on redjujube images using the hybrid wavelet transform and to detect the defects on red jujubesurface. With the help of existed algorithm model, this paper finished research works asfollowing:(1) To improve the performance of the hybrid wavelet transform, the DualTree-Complex Wavelet Transform was selected as the male parent among SUREShrink,BayesShrink, AdaptBayesShrink, as well as the Dual Tree-Complex Wavelet Transform. Thetotal variance was selected as the female parent among median filter, average filter, Wienerfilter and total variance. To reduce the block effect in denoised images, a two-dimensionalencoding operator was proposed. Therefore, two methods were hybridized using2D encodingoperator. The visual quality of these denoised images was improved, and both the peaksignal-to-noise ratio and the computing efficiency were increased. The experimental resultsshow that the improved algorithm in this paper can achieve better denoising performance thanall of the other methods.(2) The real noise on red jujube images such as the dust was studied in this paper. It was pointed out that the dust noise was biological. It meant that the dust noise would changewhen the temperature, the moisture and time changed. The denoising performance achievedby the proposed algorithm was better than all of the single methods, such as the mean filter,Wiener filter, as well as the DT-CWT, and the existed hybrid wavelet-based method.(3) The flaw is the most common and serious defect on red jujube surface. Comparedwith other types of defects, the flaw is often long and narrow, thin and small, thus it isdifficult for general methods to detect it effectively. Therefore, this paper proposed a flawdetection method based on the wavelet multi-resolution analysis combined with the gradientvector. The wavelet transform can focus on any details of an image. And the gradient vectorof an image is stable when the illumination changes. This method made good use of the goodcharacters of the wavelet transform and the gradient vector. The method proposed in thispaper can detect the flaw defect on red jujube, although the size and the shape of the flaw aredifferent, and the position of the flaw is various. The experimental results shows that theperformance of the propose method is superior to some conventional methods, such as theOtsu method, the region grow method, the Canny and the Roberts edge detection operators.