Microscopic image impulse noise filtering of Chinese herbal medicine using pulse coupled neural networks and morphology

Abstract

In order to filter out microscopic image of Chinese herbal medicines (CHM) which is exposed by impulse noise pollution in the process of collection and access, and improve the efficiency and accuracy of the images in the follow-up detection and identification process, an algorithm of impulse noise detection and the two-step filtering using improved Pulse Coupled Neural Networks (PCNN) and morphology is put forward. First, it is identified that the location of impulsive noise in microscopic images of Chinese herbal medicines according to the characteristics of PCNN model, and then the first step processing is adaptively adopted with increase noise and image noise point neighborhood information for the images, finally, the next step filtering is processed with the mathematical morphology which is able to better protect the edge of the details. The theoretical analysis and experimental results show that the algorithm can self-set the detection threshold, miss less mistakes noise detection, and has highly detection accuracy. It can effectively filter out impulse noise, especially to the high noisy polluted images. The method is not only objectively superior to Wiener filtering, median filtering and morphological filtering in index evaluation like power signal-to-noise ratio (PSNR), the mean square error (MSE), the resistance to noise ratio improvement factor (SIF), but also subjectively improved much in the visual effect.