The most significant reduction of execution time is achieved in direct, partial rotation algorithm. The partial rotation is about ten times faster than standard indirect rotation. The algorithm can be still improved by resignation of rotation of RGB image and make all calculations with usage of known angle.

Another optimisation was reduction of Hough lines from near to 100 to 5-10 by use high threshold and requirement of high number of neighbours. Important influence to the execution time has also analysis of the position of a banknote just in vertical direction.

Otsu algorithm also is some way of improvement of effectiveness. It calculates the most optimal threshold for every image.

One of the ideas for the future is to improve the algorithm of detection of contours of the flag (European flag in the left side of the bill), which unfortunately is not enough improved and its results now are not quite significantly in order to exploit it. However, the algorithm of flag localisation gives a way to find the position of the banknote easier and more direct to, The algorithm just needs to check the position of very small region of European flag. In this way it would be also possible to recognise multiple banknotes or banknotes with complex background, that would be really huge improvement.

In general algorithms and programs for image manipulation and image processing are kind of "Never ending story" and always there is something to improve. As it was told at the first chapter of this paper: every image is eternal world with different requirements and characteristics. Can be said that the creation of the program that would be able to recognise the complex pattern in 100% in all possible images, that are possible to recognize by a human, is extremely difficult, but who knows, maybe one day the computer vision would arrive to the level of intelligence of human vision and brain.