In this project we find the characters present in the captcha images. First pre-process the image byseparating the background from the foreground characters and after segmenting the given image into individual characters, fed the resulting characters into a custom convolutional neural network(CNN) for classification. Owing to the different orientations of the various characters in the image and different backgrounds, a powerful non-linear model was conspicuously our first choice and it proved to do well on the given dataset as well.

The project report can be found here