Image recognition Android

Question

For my bachelor thesis I have to make an app that recognize logos, eg : I see the logo of a car and I want to find out what car it is. I take a picture of the car’s logo and the app should recognize the image and send me back the Word „Mercedes” so that I can search for information about the car online. I would like that no matter the position of the logo or the light or the color, the app to recognize what that logo represents.

I have tried with recognize.im API but it doesn't work well because it is callibrated and adjusted for comparison not classification, and I definitely need classification.

I would like to go with on-cloud recognition, but on-device would work too (in which case what algorithm should I use ?).

Thank you very much

Answer 1

There are few APIs which provide general image recognition such as Google Vision or Imagga. These services can give you some general information about the scene - for example it can tell whether there is a car in the image or not.

However, your car brand recognition task is very specific and you might achieve better results by using customizable service such as vize.ai, which allows you to train your task-specific API endpoint. In order to train it, you need to prepare example images with logos for each car brand you want to recognize (30 to 50 images per brand). You upload these images to vize.it using the web browser interface and you'll get the API endpoint to a classifier trained for your task. Then you can simply classify new images by calling the API.

Edit:More details added (as requested by ρяσѕρєя K)

Disclaimer: I'm working at vize.it. Edit: Link changed