Universal Currency Converter Application
This paper proposes a method of inspecting water mark on currency note by using correlation mapping and backpropagation neural network. In this method, the location of water mark is detected by correlation mapping with the edge on reference image. To certify the water mark, the edge information from the shadow of water mark is inputted to backpropagation neural network and it is classified into the currency note or the copy. In the experiment, five samples each of five types (B20,B50,B100,B500,B1000) of Thai currency note were trained, and 20 samples of each were tested. The results reveal that the currency notes are inspected approximately with 99.00%, accuracy of recognizable type of currency note and 100.00% by using all of the edge information of currency note and the copies were rejected.
Nowadays, there are a lot of spurious currency notes. Since highly efficient technique is required to
make water mark on currency note and water mark can be seen only by its shadow, the water mark is very crucial information to inspect spurious currency note. Many researches involve in currency note
recognition but there is no any research which directly related in inspection of water mark, such as Angelo Frosini, Marco Gori and Paolo Priami [ l ] presented optoelectronic technique which is used for sensing signal, and the reflective signal is inputted to autoassociator neural network for classification of currency note. Minorii Fukumi, Sigeru Omatu, Fumiaki Takeda and Toshihisa Kosaka proposed rotation- invariant in coin recognition system by creating circular array and their gray level values were feed into neural network Yegnanarayana proposed a method of unsupervised
texture classification by vector quantization and hopfield neural network.Suwat, Pramote and Kosin proposed classification of thai note and recognition system by two frequency bands and fieedfonvard neural network.
The system hardware is shown figurel. It consists of CCD camera as an input unit of currency note (I), A/D converter and 512*512 pixel frame card memory with 8 bit 256 gray level for converting analog signal to digital and save to card memory for next processing (2), microcomputer for processing the image in card memory (3), monitor for showing the results (4), and light source for shining the note ( 5 ) . The system software flowchart is shown in figure2.k can be described, the location of the water mark is detected (1). The edge information from the shadow of water mark is derived from shining
image, and it is inputted to neural network for certifying (2). The edge information all of currency note is inputted to neural network for checking