Description

This repo contains all 4 requested homeworks of the course Computer Vision of the 7th semester.

Inside each homework directory there is a README file where you can find a quick glance of the corresponding results.

• This repo is for demonstration purposes only

HW1 - Document Image Analysis

• Sub-text areas detection with bounding boxes.
• Information measurements of each sub-text area.

HW2 - Panorama Image Stitching

• Using SIFT and SURF descriptors.
• Using Image Composite Editor.
• Compare the results.

HW3 - Multi-Classs Image Classification using BOVW

• Creation of a visual vocabulary based on the Bag Of Visual Words model using the K-Means algorithm.
• Extraction of BOVW Descriptor for every image of the training dataset.
• Using classifiers k-nn and SVM (with the confidence criterion of One vs All).
• Classification assessment using test set and check of the effects of number k of neigbours and kernel type of SVM.

HW4 - Multi-Classs Image Classification using CNN in Keras-Tensorflow

• Using a non-pretrained NN.
• Using a pretrained MobileNet.
• Classification assessment of both architectures using validation-test set.

Image Datasets

• HW1: Scanned documents with and without noise (salt and peper).
• HW2: Custom images.
• HW3: Caltech 256 Image Dataset
• HW4: BelgiumTS - Belgian Traffic Sign Dataset