Automating Joiners

by Natalie Katz and Neta Zeira Supervised by Gur Harary

Abstract

This paper describes Automating Joiners, a process which automates the creation of an image combined from several different pictures. This set of pictures may be taken from a dynamic scene from a fixed angle or from a stationary scene from changing angles. Within the project the emphasis was making a beautiful picture, seeking an artistic outcome. The different stages of the algorithm, described in detail further on, started with feature extraction using an outsourced SIFT algorithm. Thus it was possible to find the compatibility between each and every picture within the set. Once the compatibility was extracted, the pictures were transformed to a single set of coordinates through a detailed process of grading the possible transformations. Once all in the same coordinates the final image frame was calculated, while keeping in mind the need for reliability, meaning having enough data to work with further on. At this stage object detection algorithms, a Viola-Jones inspired face detection algorithm and a saliency algorithm, were used for the benefit of grading different sections of each transformed image. These grades were utilized to screen the less useful sections. Finally, sections from different pictures were randomly arranged into the final image frame. The final product was chosen from the above mentioned arrangements by reviewing the consecutiveness between the various sections.

Tools

This project was implemented in a Matlab environment, with the help of a SIFT algorithm, a Viola-Jones based face detection algorithm and a saliency algorithm.

RELATED DOCUMENTATION

• Project report [MS Word - 2.58 MB]
• Powerpoint slides [MS Powerpoint 2.71 MB]
• Project poster [Adobe Document 2 MB]
• Source Code [Zip]

Figure 1 - An example for the outcome of the algorithm.