Abstract:
SketchIT is a tool that recognizes hand drawn objects and fetches natural images of the recognized objects.
The project aims to recognize sketches drawn by hand in different orientations and spaces on an android application that would be recognized by an android application, identified and the natural images of the recognized object displayed.
The problem lies in the correct recognition of the hand drawn image. The sketches due to their different angles and the perspectives of the user are drawn in different ways, which are hard to identify. One may draw it lateral, horizontal, vertical or whatever angle one might want. We have improved, through different data augmentation techniques, the trainer, to recognize those sketches.
The importance of the project relies on the focus to accurately recognize the drawn sketch through different techniques. Since the drawn sketch would depend on the perspective of the user, the algorithm would analyze the sketch on different parameters, accurately recognize the sketch, match it with the images and database, and label it a tag. It would then, retrieve relying on the tags, the images of the recognized image from the database and display them. The user would be shown natural images of the drawn object.
Through machine learning and artificial intelligence, the sketches are tailored and recognized. We have fed the trainer with over 200,000 images.
We have applied customized Sketch-a-Net on our modified dataset, which gives an accuracy of around 70 percent approx. We hope to increase it further in the near future.
