The document describes a system for recognizing numbers on LEGO packages from images. It uses image preprocessing techniques like HSV color thresholding and affine transforms to detect the LEGO logo and standardize the image. Optical character recognition is performed using KNN, SVM and Random Forest classifiers on Canny and gradient features of cropped number regions. A convolutional neural network is also trained on labeled image data to recognize numbers, achieving over 95% testing accuracy. The system provides an automated solution for LEGO package recognition but further improvements are needed for accuracy and mobile compatibility.