Removing Noise

[Document/Medical only] The DespeckleCommand class is designed specifically for removing specks from 1-bit images, such as FAX transmissions or scanned documents.

MedianCommand class is an effective general method for removing salt-and-pepper noise from any kind of image. You control the strength of the method by specifying the size of the neighborhood (the surrounding pixels used for calculating the median value). This method causes minimal blurring of the image.

The AverageCommand class can be used to remove uniform and Gaussian noise, but there is more blurring of the image than with the median filter. You control the strength of the class by specifying the size of the neighborhood (the surrounding pixels used for calculating the arithmetic mean).

The AddCommand class can be used to average the images in a list, eliminating random noise contained in the images.

The AddWeightedCommand class class can be used to average the images in a list, eliminating random noise contained in the images. Typically, you would call this class for a series of images taken for the same object at short intervals. This class can eliminate random noise by performing a weighted average.

The GaussianCommand class smooths or blurs an image by applying a Gaussian filter to each pixel of the image. The amount of blur is determined by the size of the neighborhood used by the class.

The RemoveSaltPepperCommand class removes the salt and pepper noise from the specified image.

[Document/Medical only] The binary filters (for erosion and dilation of black objects) can be used for noise removal. The BinaryFilterCommand class applies directional binary filters. The MaximumCommand and MinimumCommand classes let you control the neighborhood size for erosion or dilation. One noise removal technique with these methods is to create two copies of an image, apply a dilation filter to one of them and an erosion filter to the other, then combine the two using the CombineCommand class with the OperationAverage (averaging) flag.

[Document/Medical only] The AutoBinaryCommand class transforms the image into a binary image using a threshold calculated automatically based on statistical features of the image. Best results are obtained with grayscale images.

The DeinterlaceCommand class deinterlaces video source images by removing the black lines. It proceeds by merging lines in the image together and/or blending them together.

The SmoothEdgesCommand class smooths the edges of an image.

The DynamicBinaryCommand class converts an image into a black and white image without changing its bits per pixel.

Fourier Transforms

Fourier Transforms are useful in removing harmonic noise like:

• Herringbone patterns seen in video images
• Sine wave patterns
• Moiré patterns
• Halftone patterns
• Interference patterns

Fourier Transforms are also useful in removing noise from video signals and CCDs.

Fourier transforms convert an image from one where intensity varies over space into one where the intensity varies with frequency. After an image has been described (transformed) as a series of frequencies and phase, that information can then be analyzed using a power spectrum. The power spectrum is a two-dimensional map that plots the frequency of pixel intensity values. Low frequencies lie close to the origin and high frequencies close to the edges. From the analysis suitable filters can be applied to remove those parts of the image that correspond to the noise in the image. The image can then be transformed back into the original image, without the noise, using the inverse command flag.

A Discrete Fourier Transform is a Fourier transform that uses a series which has values that have both a real and an imaginary component. Discrete Fourier transforms reveal periodicities in the data as well as the relative strengths of those periodicities.

Discrete Fourier Transforms can take a long time to compute. To reduce computing time, Fast Fourier Transforms can be used, which reduce the number of computations from 2N² to 2N Lg N. However, using a Fast Fourier Transform requires that the number of points in the series be a power of 2. The image can be transformed back into the original image, without the noise, using the inverse command flag

The following classes provide Fourier transform capabilities:

[Document/Medical only] The AutoBinarizeCommand class applies binary segmentation to a bitmap automatically. Pixels are compared to the threshold value. If the intensity of the pixel is higher (brighter) than the threshold value, the pixel is set to white. If the intensity of the pixel is lower (darker) than the threshold value, the pixel is set to black. This is useful for improving recognition results (OCR, Barcode, OMR, ICR).

The AnisotropicDiffusionCommand class Applies a 2-Dimensional Anisotropic Diffusion filter on an image to reduce varied patterned and non-patterned noise. Reduces noise and preserves edges better than the Median or Gaussian filters.

The TADAnisotropicDiffusionCommand class is an iterative filter that performs tensor-guided anisotropic diffusion in order to reduce noise while preserving the edges in the image. Reduces noise and preserves edges better than the Median or Gaussian filters.

The SRADAnisotropicDiffusionCommand class applies a 2-Dimensional Anisotropic Diffusion filter on an image in order to reduce noise and speckling. Reduces noise and preserves edges better than the Median or Gaussian filters.