Acceleration of Server-side Image Processing by Client-side Pre-processing in Web Application Environment

The goal of this paper is to verify whether the use of client-side TensorFlow.js, a WebGL-accelerated JavaScript library for machine learning, can accelerate the processing of common photos by computer vision cloud services, such as detection and recognition of specific features like age, sex, expression or specific people in the image. This acceleration is based on pre-processing the input image, namely detecting human faces, which greatly changes the amount of input data that need to be uploaded to the cloud service and thus the amount of uploaded data compared to the original photograph. The upload speed of Internet connection often is, in the case of computer vision cloud services, the bottleneck of the whole system. That's why decreasing the amount of uploaded data in time shorter than the difference between the total of upload and cloud service processing time of the original and the pre-processed image leads to acceleration.