Representation research based on sketch images
XIAOZHOU ZHOU, Minhui Jin, Yinqing Zhang · 5th International Conference on Computer Information Science and Application Technology (CISAT 2022) · 2022
In modern computer science, tasks such as optical character recognition are of great importance. Yet, for artificial neural networks it’s never easy to distinguish what a scrawled image indicates.In December 2016, Google post a drawing game which would randomly pick up a word and ask the player to draw an outline within 20 seconds. An AI was used to judge if your work looks similar to what you were asked to draw. With more and more people playing this game, Google has collected over 1 billion drawings.Freehand sketching is a dynamic process where points are sequentially sampled and grouped as strokes for sketch acquisition on electronic devices. While most existing models simply convert the sketch vectors into binary images and pay little attention to the time information, we organize training data and adapt traditional image recognition networks in a novel way. In this report, we will introduce our solution to this doodle recognition problem, including data pre-processing and image recognition. We would also explain experiment process and the main challenges we met in detail.