Use of Neural Network Models Supported by Biological Visual Edge Detection in Image Processing

The research attempt aims to apply the human visual system biologically inspired edge detection mechanisms, that were observed specifically in the primary visual cortex (V1), to neural network models, for better image processing and classification. The suggested procedure for the simulation of the edge detection mechanism of V1 is a model that takes greyscale images and their corresponding Canny edge maps as dual-channel inputs, and thus, the V1 function of edge detection is simulated. The objective of this provision is the recognition of specific objects on images through a process that is noise and structure complex without compromising the performance of the system. The proposed method was verified experimentally using the CIFAR-10 dataset, and the outcome was that the accuracy of classification was significantly improved, 9.5% as compared to the standard single-channel input approach of convolutional neural networks. The new model demonstrated higher strength in the discovery of object outlines and distinguishing the most prominent attributes in the case of complex visual scenarios. The results of this study have been given enough evidence that shows how models inspired by nature can impact and bring improvements to machine learning algorithms that handle image classification. Still, the current research can develop human-made intelligence based on nature, and the importance in the next-generation visual computer model design is the duplication of the visual perception principles of nature.