Television image before and after processing
Television station identification logos have become an increasing source of annoyance to the viewer in recent years. The recent relaunch of SBS Two, with its huge annoying new logo, prompted speculation as to how these logos could be removed from the picture.
Most free-to-air stations employ a semitransparent, or alpha-blended, logo overlayed onto the picture content. It occurred to me that all the information necessary to reconstruct the original picture is being transmitted. Each pixel where the logo is visible can be regarded as a transformation function (red trace on the graph) applied to the original, unmodified image. If the inverses of all of these functions are found (blue trace), and the appropriate one is applied to each pixel in the transmitted image affected by the logo, the original scene should result, with the logo removed.
Function and its inverse. (From Wikipedia)
The ideal implementation of this concept would be a small module that could be connected inline with the video signal to the display device. The video signal would be passed through, but with the station logo removed. Facilities would be provided for 'learning' the appearance of the logos during initial setup, or after a change made by the TV network. The user interface for this process should be as simple as possible. The module would automatically recognise which logo is present on the incoming signal, out of all those logos known to it, in order to cater for different channels. This function would also detect when a video source without a superimposed logo was selected, and inhibit any cancellation functions.
In order to achieve this, a custom hardware design would be required. However, in order to prove the algorithms before committing to the design and manufacture of a circuit and PCB, a software-based approach was used initially.
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.