Band-Pass Filtering for Non-Stationary Noise in Agricultural Images to Pest Control Based On Adaptive Semantic Modeling

摘要

Image analysis has been used in a very large scale for different purposes. When an image is captured by a digital sensor, it is usually affected by some type of noise, even the smoothest ones. Therefore, image enhancement and denoising process are important tasks of digital image processing. This paper presents an algorithm to reduce non-stationary noise with the combination of a Low-Pass Filter (LPF) and a High-Pass Filter (HPF), in conjunction with an adaptive semantic model. To simulate the usefulness of such arrangement, a non-stationary Gaussian noise has been applied to an image, which has been splitted into the four quadrants, all of them having the same dimensions. In fact, such a noise with different intensities, has been added to the image in each of its quadrants. The Peak Signal-to-Noise Ratio (PSNR) has been used to measure the best cutoff frequencies for both filters, as well as rules based on semantic concepts have been structured for decision making. Furthermore, for the validation of the algorithm we have taken into account the evaluation of the Mean Squared Error (MSE) using a typical digital image obtained from a crop of maize with the presence of the earwornm (Helicoverpa Zea). Besides, the denoising process demonstrates the efficiency and the satisfactory performance for the non-stationary noise filtering in agricultural images.