A Survey on Breaking Technique of Text-Based CAPTCHA

Table 5

Comparisons of common denoising methods.

Denoising method	Typical algorithm	Implementation	Advantages	Disadvantages
Denoising method based on filter in the spatial domain	Average filter	The gray value of pixel is replaced by the mean of its neighboring pixels gray values.	The irrelevant details and gaps are removed.	The image is blurred.
	Median filter	The gray value of pixel is replaced by the median of its neighboring pixels gray values.	Remove effectively the salt and pepper noise, speckle noise.	Not applied to the image with many dots, lines, and spires.
	Wiener filter	The minimum mean square error criterion is used to adjust the filter effect.	Remove effectively the Gaussian noises.	Computation is complex.
Denoising method based on Gibbs and Hough transform	Gibbs	Markov random field theory.	Remove effectively noise points.	Not applied to irregular interference line.
	Hough transform	The straight line in the image is detected by using the point line duality of image space and Hough parameter space.	Remove effectively interference lines.
Denoising method based on morphology	Open operation	First corrosion to expansion.	Smooth contours, cut off narrow lines, and eliminate fine.	The effect of denoising varies with operation mode and the size and shape of structural elements; the experiment needs to be repeated; the adaptability is poor.
	Close operation	First expansion to corrosion.	Smooth contour and fill holes, gaps, and fracture of contour line.
Denoising method based on connected component	Connected component	The recursive method is used to find the connected domain to deal with pixel points, and then denoising based on gray features and morphological features of connected domain.	Remove effectively the noise interference, and the original details of the characters are generally not lost.	Need to analyze character’s properties; hard to determine distinguish features.
Denoising method based on wavelet transform	Wavelet transform	Find the best mapping of original image in the wavelet transform domain to restore the original image.	Retain more image details.	Complex computation and it needs to adjust relative parameters.