Banknote localisation

Banknote localization is the most important part of all banknote recognition process. All other algorithms depend on the result of this part. First step of localisation algorithm is to find contours of the image and then try to decrease noise without loosing important details. To find contours Canny algorithm, that is one of the most efficient algorithm for contour detection, is used.

To find the object is necessary to calculate histograms of sum of pixels in horizontal and vertical direction. In this paper horizontal histogram is the array of sums for columns (vertical rows), horizontal is the axe of arguments (arguments from 0 until width of the image). There is similar definition for vertical histogram: it is the array of sums of horizontal rows for vertical axe of arguments (arguments from 0 until height of the image).

On every histogram can be found the maximum value (the position, where are more white pixels, maximal sum). Then, there is marked the region where can be found pixels that are greater than 10% of maximum. First looking for object is done with small margin added to calculated area for minimizing noises. Then the image is cut just to the interested region. After it the algorithm starts second time in previously found region and is called without margin, so returned just the most interested and significant object. In this way the second call can find exact position of the bill.