Recently two Israeli scientists published a screen capture video of their experimental image resizing tool in action. Tool uses a seam carving algorithm to identify and delete areas of the picture most unlikely to include important information. The demonstration has rightfully gained quite a lot of attention. Inspirational video has already spawned multiple different tools, plugins and web services that attempt to reproduce the same functionality on different platforms. While seam carving may even in itself be revolutionary it is also a gateway to even more impressive new tools. One such idea is combining the algorithm with golden section rules.
Golden section, rule of thirds or golden mean are all different names for the most commonly used rules of thumb in image composition. Variety of different software tools have been developed to help using these rules, but none have really gained wide acceptance. The biggest reason has been that the tools haven’t really been integrated into the creation process and the user interface has been too clumsy. Artistically oriented people tend to shy away from using such rigid tools and trust their gut feeling instead. Consequently a tool that would have a chance at catching on would need to be very well integrated into designers’ creative workflow.
This brings us back to resizing and seam carving projects. One of the most common tasks in graphic design has always been resizing elements on a page. At the same time the seam carving demos all have the limitation that setting the image proportions is quite inconvenient. Now if we were to integrate a smart golden mean resizing tool with seam carving solid compositions could be reached. The goal of such work would be a self-explanatory tool that all graphic designers would be willing to embrace.
The first steps toward an user-friendly tool have already been taken. Automata for example is an experimental software project that uses the ancient rules to help designing web layouts. The same algorithms used for calculating the web page element dimensions could just as well be used to find optimal proportions for elements in images.
This same algorithm could come in quite useful in video editing work as well. Of course seam carving algorithm is quite computationally heavy and rendering times would with current processors be long. The capacity grows all the time and it could be that in the distant future we could render in real-time and even use the same algorithm inside a camcorder helping the human operator compose augmented reality in real time. Human form recognition could be used to mask parts of the image that are not wanted to be changed and these same positioned to the background.