Project consist in the development of an AI model that can be able to recognize handwritten digits and characters of a socioeconomic survey Link.

In order to achive the highest possible accuracy, we use a convolutional neural network and we take advantage of shape detection.

To train our models we've used the NIST Special Database 19.

We've implemented using Tensor Flow framework in python.

Table below shows performance of several convNet models we've built:

Project consist of three main directories:

• Extraction - Contain the scripts that extract individual characters of the survey form. ** Detect page, ** Feature extractor

• Modeling - Contains convnet models designed.

• Api - Script that integrates the above. ** Engine