Projects

My course projects (some of which I have participated as a team member) can be found below. These projects are given as assingments and some of them may be incomplete. They are presented here for demonstration purposes. For my master thesis you can see the research section.

3D Home Architecture Program

with

This is a fully functional 3D home architecture and decoration program, developed as a course project for the computer graphics course. It uses OpenGL and is written in a completely object oriented manner.

The project consists of three parts:

First, the user creates the base of his house using constructive solid geometry. The first video to the right demonstrates the interface of the first part and the use of constructive solid geometry. The user creates, rotates, adds and subtracts rectangles in order to create the base of his house. He can zoom in, zoom out, pan and hide or show grid lines while doing so.

After completing the base, the user converts it to an empty 3D house. The second part of the program allows the creation of inner walls, creation of doors and windows, addition of furniture and light sources and the assignment of different textures to walls and the floor. 3D Studio Max models (3DS files) can be imported and used as furniture.

While constructing and decorating the house, the user can select between a 4-viewpoint or a single enlarged viewpoint interfaces. All of the viewpoints can be selected and changed to be a perspective, isometric, oblique or a conceptual view. The user can zoom in, zoom out, pan and rotate the views.

The second and the third videos to the right demonstrate some aspects of the second part of the project.

After constructing and decorating the house, the user can wander in and around his house in a day or night setting seen from a first person perspective. The last video to the right shows the third part of the project in a day setting.

Superresolution

with

This is the Video Processing course project. It takes as input multiple low resolution images and outputs an image having four or sixteen times the resolution as the input images. It only works for input images having only global motion.

The method we have used is the one proposed in Fast and Robust Multiframe Super Resolution by Milanfar, P. et al. The journal paper can be seen here.

To detect the global motion between the low resolution images, we use the phase correlation method. In order to obtain sub-pixel accuracy, we first upsample the low resolution images.

We have tested our project with two different data sets.

The first data set is created by taking a photo, applying different amounts of vertical and horizontal shift to it and downsampling the shifted versions using the nearest neighbour method. The original high resolution image, one of the downsampled images and the result of supperresolution can be seen by clicking the first image to the right. For comparison purposes, the upsampled version of the low resolution image using bicubic interpolation is also given.

The results are very good but these results can be deceptive because in a real-life scenario the low resolution images are generally are not downsampled using the nearest neighbour method.

In order to obtain more realistic results, we have formed a second data set. This time we took pictures of a scene while slowly shaking our hands. Then we downsampled these images using the bilinear interpolation method.

The results can be seen by clicking the second image to the right. This time the composed image is not as crisp as the previous one; however, still the difference between the result of superresolution and the result of upsampling using bicubic interpolation is quite apparent.

SVN vs. FFN

with Murat Deniz Aykın

This is a small applet which compares the performances of Support Vector Machines and Multilayer Perceptron Artificial Neural Networks.

...

...

This site is created by Eren HALICI. If you are using IE to view this page, don't.