Extracting Barcodes from a Extracting Barcodes from a Camera-Shaken Image on Camera-Shaken Image on

Camera PhonesCamera Phones

Graduate Institute of Communication EngineeringNational Taiwan University

Chung-Hua Chu, De-Nian Yang and Ming-Syan Chen

報告者 : M97G0225 黃庭筠

OUTLINEOUTLINE

• INTRODUCTION

• BARCODE EXTRACTION ALGORITHM

• EXPERIMENTAL RESULTS

• CONCLUSION

INTRODUCTIONINTRODUCTION

• problem is to extract and restore the barcode to improve the correctness of the recognition.

• Extracting Camera-Shaken Barcode (ECSB) algorithm for the problem

BARCODE EXTRACTION BARCODE EXTRACTION ALGORITHMALGORITHM

Barcode area extractionBarcode area extraction

ECSB first uses Edge Detection to detect the rough barcode area as depicted.

Then, ECSB dilates the edge-detected image with two directions (i.e., vertical and horizontal) such that we can obtain the precise barcode area as shown.

Barcode RestorationBarcode Restoration

Use a large constant c to approximate an unknown parameter SNR (Signal to Noise Ratio), where c is larger than or equal to the height of the distorted image.

Barcode RestorationBarcode Restoration

Barcode RestorationBarcode Restoration

g(i, j) = f(i, j) * h(i, j) + n(i, j)

g(i, j) : presented by a linear system of a convolution

f (i, j) : denotes the original image

n(i, j) : defined as a Gaussian white noise with zero mean

h(i, j) : (PSF) can be viewed as the filter caused by the

shake

Barcode RestorationBarcode Restoration

N : denotes the length of the camera motion

Θ: denotes the direction of the camera motion

Camera-Shaken Direction and Camera-Shaken Direction and Length EstimationLength Estimation

• estimate the camera-shaken direction Θ :

1. first transform the camera-shaken image to a

Cepstrum-domain image C (i, j)

C (i, j) : Inverse Fourier transform of log(1 + G (i, j) )

G (i, j) : is the Fourier transform of g (i, j)

2. use Hough transform to find out the camera-shaken

direction Θ in C (i, j)

Camera-Shaken Direction and Camera-Shaken Direction and Length EstimationLength Estimation

• camera-shaken length N :

1. calculate the average of all pixels in C' (i, j)

2. Finally, we calculate the first zero crossing where the

camera-shaken length N

Camera-Shaken Direction and Camera-Shaken Direction and Length EstimationLength Estimation

Camera Shake RestorationCamera Shake Restoration

Use Wiener filter to determine the solution F (i, j)

Sf : power spectrum of the original image

Sn : power spectrum of the noise

H (i, j) : frequency domain of h (i , j)

H*(i, j) : complex conjugate of H (i, j)

∣ H (i, j) ^2 : H*(i, j) H(i, j)∣

EXPERIMENTAL RESULTSEXPERIMENTAL RESULTS

EXPERIMENTAL RESULTSEXPERIMENTAL RESULTS

• Tag Richardson Lucy method (TRL) :

Uses tag-based identification to extract QR code,

and then Richardson Lucy method (RL) to restore

the extracted QR code.

EXPERIMENTAL RESULTSEXPERIMENTAL RESULTS

• root mean-square error (RMSE)

H : height of f (i, j)

W : width of f (i, j)

EXPERIMENTAL RESULTSEXPERIMENTAL RESULTS

CONCLUSIONCONCLUSION

• In this paper, we proposed an efficient algorithm to extract the 2D barcodes in a camera-shaken image.

• The experimental results have showed that our approach is not only of smaller running time but of higher accuracy of the barcode recognition in a mobile information environment.