Image Processing Course - Final Project 1998
Image Mosaics
Objective
We have implemented a Mosaic generator software based on the
paper "Image Mosaics" by Adam Finkelstein and Marisa Range, published in RIDT'98 Proceedings, St. Malo, France.
Abstract
The paper presents a method to create image mosaics, a collection of
small images arranged together in a way that, from a distance, they suggest a larger image.
Team
Slide presentation
[ Acrobat (PDF) version ] [ Postscript (gziped) version ]
Description
The algorithm basic idea is: given an image (target image) create another
image (mosaic) build up from several smaller images (tile images).
The objective is that the mosaic image ressambles the target image.
The Mosaic generator pipeline consists on the following steps:
- Choosing Images:
Choose a target image and a set of images from where the tile images
will be taken. The mosaic can also be made of a unique repeated tile image.
- Defining Tiling:
Define the spatial partitioning of the target image in small rectangular areas.
- Image arrangement:
For each rectangular area of the target image, determine a tile image that better suggest
the content of that area.
The concept of "better suggest" may vary. We may say that an image suggests an other when their
average colors are almost equal. Another way is to compare features.
- Color Correction:
Alter the colors and/or the luminance of each tile image matching the colors
of the region in the target image while preserving features.
Implementation
The software was implemented in C++ using the FLTK user interface library in Windows platform.
We have implemented two different methods of image search: one that searches similar features using
the Haar wavelet and another that searches the database for closest avarege colors.
We also developed two different color correction schemes: one that alter the overall chrominance of the tile to better match
the original tile and another that changes the luminance, pixel per pixel, to the equivalent in the original image.
Downloadable archives:
Software
Sun Landscapes database
Vangogh database
Results
Rio panorama with Haar search and chrominance correction.
Mona Lisa with average color search: chrominance correction (left) and luminance correction (right)
Van Gogh with Haar search and chrominance correction
Van Gogh with average color search without color correction
Van Gogh with average color search and luminance correction