부산대학교 도서관

The noise would be a significant element that affects the quality of leaf images. The level of valuable features that could be extracted from the image has frequently been reduced by the level of noise, also some essential image sections are most often distorted. Image noise has been experimented with by several analysts as the spontaneous variance of illumination or color details within images leads to noises while acquiring. Noise in a leaf image has been the outcome of different forms of errors induced by multiple causes such as the atmosphere and also the instruments involved and is added as a result of errors that arise during processing the image, encoding, and storing. Mainly the effect of Gaussian-Noise (GN) induces higher or lower contrast in both the edge region of the input image that degrades the quality of the leaf images. This research article discusses the strategies and procedures for removing noise from leaf images. The primary objective here would be to upgrade the quality of the leaf image by preprocessing for improving the performance of the automated Leaf Disease Detection (LDD) model. In this research, we propose the following filtering techniques for preprocessing the leaf image “Discrete-Cosine-Transform (DCT)”, “Discrete-Wavelet-Transform (DWT)”, and “K-means Singular-Value-Decomposition and DWT (K-SVD-DWT)”. The superior filtering approach was determined using the metric “Peak-Signal-to-Noise-Ratio (PSNR)”. The outcome of the highest PSNR denoised image can be transmitted into the segmentation task for further LDD process.