The impact of input data alignment on the neural network training process

In this article, the author presents a problem resulting from inaccurate input data alignment in supervised neural network training processes. The neural network is tasked with generating a finite impulse response filter designed to improve the quality of digital images acquired from image acquisition devices based on CCD sensors. To solve the problem, the author presents a method based on finding the axis and angle using the correlation coefficient.The task of digital image acquisition devices is to convert an image from analog to digital form. A simplified version of the digital image acquisition path is shown in Figure 1. (…)

The impact of input data alignment on the neural network training process – PDF file for download