Sparse noise minimization in image classification using Genetic Algorithm and DenseNet

Abstract

Noise handling is a critical aspect of image processing, which can significantly affect the accuracy of classification and recognition algorithms. In this paper, we propose a technique for improved noise handling in sparse input feature maps where the noise signal is also sparse. The signal-noise relationship is formulated as an optimization problem which is solved by a genetic algorithm. The genetic algorithm is applied to optimize the setting of a non-convexity parameter which yields a more accurate image sparse matrix. The resulting feature map is then classified using a densely connected convolutional network (DenseNet). Lung computed tomography images were used for the experiments. The proposed approach achieves better performance when the classification results are compared with a case in which the input signal has not been denoised using the proposed approach.