HISTOLOGICAL IMAGE CLASSIFICATION METHODS BASED ON CONVOLUTIONAL NEURAL NETWORKS WITH GABOR FILTERS
E. A. Murin, Dmitry V. Sorokin, Andrey Serdjevich Krylov · 2024
The article presents histological image classification methods that combine the use of Gabor filters and convolutional neural networks (CNNs). Gabor filters are used to pre-process images to extract frequency characteristics and oriented texture features, improving the efficiency and accuracy of subsequent CNN processing. This approach is aimed at reducing the number of trainable parameters, increasing the learning speed and increasing the adaptive abilities of models, which is especially important when the amount of training data is limited. Various configurations of hybrid networks with Gabor filters of various sizes and their impact on model performance are investigated. Experimental results on the NCT-CRC-HE-100K dataset demonstrate that the proposed methods can achieve high classification accuracy, surpassing traditional CNNs in terms of the number of parameters and learning speed.