МАСШТАБУВАННЯ ЦИФРОВИХ ЗОБРАЖЕНЬ У ТЕЛЕКОМУНІКАЦІЙНІЙ СИСТЕМІ МЕТОДАМИ ІНТЕРПОЛЯЦІЇ ТА ЗА ДОПОМОГОЮ ЗГОРТКОВИХ НЕЙРОННИХ МЕРЕЖ

Software for scaling digital images in a telecommunication system consisting of transmission and reception subsystems has been developed. In the transmission system, the initial images are read from video cameras or graphic files, after which the image scale is reduced and transmitted through the channels of the telecommunication system. In the reception subsystem, images are read in a reduced scale, after which their scale is increased to the original. Image scaling during transmission is performed in order to reduce the load on the system channels. This is especially important for communication channels with a low transmission rate, for example, using nRF24L01 radio modules. The use of such radio modules is advisable due to their low power consumption. The image scaling program was developed in Python. In the reception system, the program is executed on the Google Colab cloud platform. Image downscaling is performed using bilinear and bicubic interpolation methods. Image upscaling is performed using bilinear and bicubic interpolation methods, after which the resulting image is refined using a convolutional neural network with a SRCNN (Super-Resolution Convolutional Neural Network) architecture. During training, fragments (patchs) of images after interpolation were fed to the inputs of the neural network, and the corresponding patchs of the original images were fed to the outputs. Parallelization of calculations was performed using a graphics processor. It is shown that the highest accuracy of the restored image is obtained when its scale is reduced using the bilinear interpolation method and increased using the bicubic interpolation method. When the scale of the original image is reduced by 2 times on the receiving side, an image with satisfactory visual quality is obtained, and the transmission time is reduced by approximately 4 times. The developed software can be used in video surveillance systems with low power consumption that do not require high data transfer rates.