All presented statistics are taken for photos with more or less plain background, without strong light reflexes and with just one object per picture. Total number of pictures used in statistics is 200.

5 Euros

Nominal	Total number of banknotes	Rotated	Horizontal position	Correct recognition	No recognition	Incorrect recognition
5 Euro	60	20	40	44	4	12
		% of Rotated	% of horizontal	% of correct recognition	% of no recognition	% of incorrect recognition
		33,33%	66,66%	73,33%	6,66%	20,00%

On a tables below can be observed that sometimes algorithm recognise false nominal for 5 Euro. Sometimes the program did not detect the nominal of 5 Euro in nominal recognition phase, then the banknote is recognised as a 20 Euro, because its grey colour is quite similar to 20 Euro blue colour. As a 10 Euro and 50 Euro the nominal of 5 Euro can be recognized when the first localisation algorithm does not return correct result. This situation can happen because the algorithm of localisation is sensitive for noise and light. Unfortunately, results of next algorithms depend on the result of localisation. It means that the number of false results can be decreased by improvement of algorithm of localisation.

10 Euros

Nominal	Total number of banknotes	Rotated	Horizontal position	Correct recognition	No recognition	Incorrect recognition
10 Euro	54	17	37	43	5	6
		% of Rotated	% of horizontal	% of correct recognition	% of no recognition	% of incorrect recognition
		31,48%	68,52%	79,63%	9,26%	11,11%

Results of 10 € are quite good. Usually the nominal is detected in nominal recognition phase. As is shown in the table below, algorithm in any case confused the 5 euros from 10 euros. It means that algorithm of nominal recognition works quite good. Some false results in this case are effects of colour recognition that every time return some result. All validations of the image localisation are made before the coulour analisis but are not so much restrict. It is because colour analisis does not need so well detected location like nominal and hologram recognition algorithms.

20 Euros

Nominal	Total number of banknotes	Rotated	Horizontal position	Correct recognition	No recognition	Incorrect recognition
20 Euro	32	11	21	27	1	4
		% of Rotated	% of horizontal	% of correct recognition	% of no recognition	% of incorrect recognition
		34,37%	65,63%	84,38%	3,13%	12,50%

20 € has the best rate of recognition that is about 84%. The reason of this situation is that colour recognition algorithm is less sensitive to the banknote in the correct position. The banknote can be moved or rotated, because algorithm just check if the colour of the object is more red or more blue. This can cause some false recognition of banknotes as is shown in statistics of 5 € and 10 €.

50 Euros

Nominal	Total number of banknotes	Rotated	Horizontal position	Correct recognition	No recognition	Incorrect recognition
50 Euro	54	22	32	41	7	6
		% of Rotated	% of horizontal	% of correct recognition	% of no recognition	% of incorrect recognition
		40,74%	59,26%	75,92%	12,96%	11,11%

50€ can be recognized in two ways: in hologram analysis or in colour recognition. Hologram analysis is very precise and detect 50 € with high probability. The problem that the algorithm needs high image quality and perfect result of localisation algorithm. Because the localisation algorithm is so much universal, a lot of times happen that only a hologram analysis is not enough. More, that the hologram is often not in adequate quality because of light reflexes. For this reason program uses additional algorithm of coulour analysis for 50 € recognition. In this way 50 € even if can not be recognise by the hologram, can be perfectly recognised by colour, even is its location is not exact.

50€ sometimes is confused with 20 €. It is because colour of the banknote can change significantly for every image. Colour depends on light, if there is any shadow and of localisation algorithm. It can happen that the localisation algorithm fails and colour recognition algorithm recognise colour of the background, not of the banknote. Unfortunately, colour is the most variable value in the image.

Global resume of statistics

Nominal	Total number of banknotes	Rotated	Horizontal position	Correct recognition	No recognition	Incorrect recognition
5 Euro	60	20	40	44	4	12
10 Euro	54	17	37	43	5	6
20 Euro	32	11	21	27	1	4
50 Euro	54	22	32	41	7	6
Totals	200	22	32	41	7	6
		% of Rotated	% of horizontal	% of correct recognition	% of no recognition	% of incorrect recognition
		40,74%	59,26%	75,92%	12,96%	11,11%

Correct result of nominal recognition for 200 banknotes arrives to 77,5%. It could be improved by increasing efficiency of the localisation algorithm and use more specific characteristic of Euro banknote (like for example European flag) to detect where exactly the bill is. In this way it is possible to decrease the number of incorrect recognitions, that nowadays is quite high. On the other hand algorithms used in the program are quite simple and very fast, there is no sophisticated operation, everything is based on histograms, that are just basic sums, so the result of so simple operations in so hide area of kind of images is quite good.