Limits...
A High Performance Banknote Recognition System Based on a One-Dimensional Visible Light Line Sensor.

Park YH, Kwon SY, Pham TD, Park KR, Jeong DS, Yoon S - Sensors (Basel) (2015)

Bottom Line: Because the banknote image is captured by the line sensor while it is moved at fast speed through the rollers inside the banknote-counting machine or ATM, misalignment, geometric distortion, and non-uniform illumination of the captured images frequently occur, which degrades the banknote recognition accuracy.To overcome these problems, we propose a new method for recognizing banknotes.The experimental results using two-fold cross-validation for 61,240 United States dollar (USD) images show that the pre-classification error rate is 0%, and the average error rate for the final recognition of the USD banknotes is 0.114%.

View Article: PubMed Central - PubMed

Affiliation: Division of Electronics and Electrical Engineering, Dongguk University, 26 Pil-dong 3-ga, Jung-gu, Seoul 100-715, Korea. fdsarew@hanafos.com.

ABSTRACT
An algorithm for recognizing banknotes is required in many fields, such as banknote-counting machines and automatic teller machines (ATM). Due to the size and cost limitations of banknote-counting machines and ATMs, the banknote image is usually captured by a one-dimensional (line) sensor instead of a conventional two-dimensional (area) sensor. Because the banknote image is captured by the line sensor while it is moved at fast speed through the rollers inside the banknote-counting machine or ATM, misalignment, geometric distortion, and non-uniform illumination of the captured images frequently occur, which degrades the banknote recognition accuracy. To overcome these problems, we propose a new method for recognizing banknotes. The experimental results using two-fold cross-validation for 61,240 United States dollar (USD) images show that the pre-classification error rate is 0%, and the average error rate for the final recognition of the USD banknotes is 0.114%.

No MeSH data available.


Related in: MedlinePlus

Recognition results for training data using the K-means method: (a) Group 1; (b) Group 2.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4507579&req=5

sensors-15-14093-f008: Recognition results for training data using the K-means method: (a) Group 1; (b) Group 2.

Mentions: In addition, we performed banknote recognition using a K-means algorithm after the pre-classification of the banknote’s direction. The features for K-means were extracted using PCA. For banknote recognition, the optimal dimensionality of the PCA was experimentally obtained in the same way as it was during the pre-classification step. Thus, the optimal dimensionality was determined to be that which produces the smallest recognition error among various possible numbers of PCA dimensions. Here, the recognition error was calculated at a 100% correct-classification rate. Correct classification means that the input banknote was correctly recognized in its corresponding class—That is, a $1 banknote was correctly recognized as belonging to the $1 class. The recognition results for the training data according to PCA dimensionality are shown in Figure 8. Using the optimal PCA dimensionality for each direction class, the image features were extracted and classified using the K-means method. As shown in Figure 5 and Figure 8, we apply PCA at the three stages such as the pre-classification of obverse (AB) and reverse (CD) sides, that of forward and backward directions (A and B, or C and D), and the banknote recognition by K-means algorithm. The optimal numbers of dimension of the feature vectors after applying PCA are different according to the PCA application for the three stages as shown in Figure 5 and Figure 8.


A High Performance Banknote Recognition System Based on a One-Dimensional Visible Light Line Sensor.

Park YH, Kwon SY, Pham TD, Park KR, Jeong DS, Yoon S - Sensors (Basel) (2015)

Recognition results for training data using the K-means method: (a) Group 1; (b) Group 2.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4507579&req=5

sensors-15-14093-f008: Recognition results for training data using the K-means method: (a) Group 1; (b) Group 2.
Mentions: In addition, we performed banknote recognition using a K-means algorithm after the pre-classification of the banknote’s direction. The features for K-means were extracted using PCA. For banknote recognition, the optimal dimensionality of the PCA was experimentally obtained in the same way as it was during the pre-classification step. Thus, the optimal dimensionality was determined to be that which produces the smallest recognition error among various possible numbers of PCA dimensions. Here, the recognition error was calculated at a 100% correct-classification rate. Correct classification means that the input banknote was correctly recognized in its corresponding class—That is, a $1 banknote was correctly recognized as belonging to the $1 class. The recognition results for the training data according to PCA dimensionality are shown in Figure 8. Using the optimal PCA dimensionality for each direction class, the image features were extracted and classified using the K-means method. As shown in Figure 5 and Figure 8, we apply PCA at the three stages such as the pre-classification of obverse (AB) and reverse (CD) sides, that of forward and backward directions (A and B, or C and D), and the banknote recognition by K-means algorithm. The optimal numbers of dimension of the feature vectors after applying PCA are different according to the PCA application for the three stages as shown in Figure 5 and Figure 8.

Bottom Line: Because the banknote image is captured by the line sensor while it is moved at fast speed through the rollers inside the banknote-counting machine or ATM, misalignment, geometric distortion, and non-uniform illumination of the captured images frequently occur, which degrades the banknote recognition accuracy.To overcome these problems, we propose a new method for recognizing banknotes.The experimental results using two-fold cross-validation for 61,240 United States dollar (USD) images show that the pre-classification error rate is 0%, and the average error rate for the final recognition of the USD banknotes is 0.114%.

View Article: PubMed Central - PubMed

Affiliation: Division of Electronics and Electrical Engineering, Dongguk University, 26 Pil-dong 3-ga, Jung-gu, Seoul 100-715, Korea. fdsarew@hanafos.com.

ABSTRACT
An algorithm for recognizing banknotes is required in many fields, such as banknote-counting machines and automatic teller machines (ATM). Due to the size and cost limitations of banknote-counting machines and ATMs, the banknote image is usually captured by a one-dimensional (line) sensor instead of a conventional two-dimensional (area) sensor. Because the banknote image is captured by the line sensor while it is moved at fast speed through the rollers inside the banknote-counting machine or ATM, misalignment, geometric distortion, and non-uniform illumination of the captured images frequently occur, which degrades the banknote recognition accuracy. To overcome these problems, we propose a new method for recognizing banknotes. The experimental results using two-fold cross-validation for 61,240 United States dollar (USD) images show that the pre-classification error rate is 0%, and the average error rate for the final recognition of the USD banknotes is 0.114%.

No MeSH data available.


Related in: MedlinePlus