PATTERN RECOGNITION

Achieving statistical dependence of the CNN response on the input data distortion for OCR problem

I. M. Janiszewski^a, V. V. Arlazarov^bcd, D. G. Slugin<sup>ba

a</sup> Federal Research Center Computer Science and Control of Russian Academy of Sciences, Moscow, Russia
^b Smart Engines Service LLC, Moscow, Russia
^c IInstitute for Information Transmission Problems of Russian Academy of Sciences, Moscow, Russia
^d Moscow Institute of Physics and Technology (State University), Moscow, Russia

Abstract: The paper proposes an approach to training a convolutional neural network using information on the level of distortion of input data. The learning process is modified with an additional layer, which is subsequently deleted, so the architecture of the original network does not change. OCR of data based on the MNIST dataset distorted with Gaussian blur using LeNet5 architecture network is considered. This approach does not have quality loss of the network and has a significant error-free zone in responses on the test data which is absent in the traditional approach to training. The responses are statistically dependent on the level of input image’s distortions and there is a presence of a strong relationship between them.

Keywords: Convolutional neural networks, pattern recognition, machine learning, distortion, Gaussian blur, OCR, MNIST.

Language: English

DOI: 10.14357/20718632190409

Bibliographic databases:

•