Polaroid Corporation

Ibrahim Hajjahmad

Munib A Wober

Michael L Reisch

Yibing Yang

Lon E Sunshine

F Richard Cottrell

F Richard Soini

John C Bowman

Disclosed herein are a method, apparatus and computer usable medium for linear filtering with filters of any desired shape and length using the wavelet transform as the computation engine by modifying the basis functions. Specifically, the method and apparatus modifies the basis function of the wavelet and/or inverse wavelet transform(s) by convolving it with the desired filter, thereby forming a modified wavelet and/or inverse wavelet transform(s). The linear filtering is performed in the signal processing and image processing related fields using the wavelet transform with a method that is mathematically equivalent to the filtering operation in the spatial domain. Linear filtering in the wavelet based domain eliminates the need for another program for software support or additional silicon area or real estate to accommodate more functions in hardware. The modified wavelet and/or inverse wavelet transform harnesses and leverages the wavelet and/or inverse wavelet basis function of the wavelet and/or inverse wavelet transform so as to perform the desired linear filtering of the digital or analog signals, without incurring extra cost in terms of software computations or hardware modifications.

Method and apparatus for performing linear filtering in wavelet based domain

In order to prevent or minimize blocking artifacts from appearing in an image due to independent processing of each overlapped block of an image by one of many different filters, true pixel correction values are calculated, then added to each pixel of the image so that the transition between adjacent blocks of pixels will be smooth. This corrective method coined 'stitching' is applied in either the spatial or the frequency domain to each block of filtered pixels in the image and generally includes the steps of:

Minimizing blocking artifacts in a filtered image

A method and apparatus are disclosed for converting frequency-coefficient matrices between a configuration in which the matrices are transforms of unoverlapped image-data matrices and a configuration in which the matrices are transforms of overlapped image-data matrices, the image-data matrices comprising image-data terms corresponding to pixels from an original image, the method comprising the steps of: deriving a conversion matrix; transposing the conversion matrix; matrix multiplying a first frequency-coefficient matrix of one configuration by the conversion matrix; matrix multiplying a second frequency-coefficient matrix of the same configuration by the transpose conversion matrix; and combining the product results to form a matrix formatted in the other configuration.

Method and apparatus for conversion of frequency-coefficient matrices

The present invention is useful in taking a visible light image for identification and other purposes without the requirement of a photobooth, regardless of the background of the visible light image. The original background of the visible light image is replaced with a preselected background. Two IR images with different intensities of IR illumination in the foreground and background regions of the scene, respectively, are compared to produce a difference image of light intensity differences between corresponding pixels of the two images. A binarized image is generated by binarizing the difference image with respect to a predetermined threshold value .theta.. A connectivity constraint is used to generate a binary mask from the binarized image, then a gray-scale mask is produced by multiplying the binary mask times a preselected modulation function of the difference image. Warping the gray-scale mask produces a transformed mask. Finally, the background replaced visible light image is generated by blending the transformed mask with the visible light image, and replacing the original background with the preselected background.

Method for replacing the background of an image

A method or system for structuring an image which corresponds to an original array of pixels, as a forward discrete even cosine transform (DCT) pyramid having a predetermined number of levels where each level is associated with a different DCT frequency band, includes, respectively, the steps or functionality of: partitioning the original array into blocks of overlapped pixels; taking a DCT of each block of overlapped pixels to generate blocks of first level DCT coefficients forming a first level of the DCT pyramid; storing the first level of the DCT pyramid; selecting and storing at least one of the first level DCT coefficients of each block of first level DCT coefficients into a first temporary array; partitioning the first temporary array into blocks of overlapped coefficients; and taking a DCT of each block of overlapped coefficients of the first temporary array to generate blocks of second level DCT coefficients forming a second level of the DCT pyramid. Additional levels can be created by repeating the previous steps, and the processed image can be restored by reversing the above sequence of steps using inverse discrete even cosine transforms.

Structuring a digital image into a DCT pyramid image representation

A method for adjusting the luminance IM.sub.HI of pixels in a high resolution image includes first forming a low resolution image of the original by grouping pixels into superpixels such as 8.times.8 blocks. A low resolution luminance correction value .delta..sub.LO is determined for each superpixel by comparing the luminance of the superpixel to both a predetermined darkness threshold T.sub.D and a predetermined brightness threshold T.sub.B. The low resolution luminance correction value .delta..sub.LO is modified to yield .delta.'.sub.LO for each superpixel by forming islands of the superpixels and smoothing .delta..sub.LO between superpixels within each island. Finally, a high resolution luminance correction value .delta..sub.HI is calculated for each pixel of the original high resolution image by comparing the luminance IM.sub.HI of each pixel to both T.sub.D and T.sub.B, keeping in mind that .delta..sub.HI is generated through a series of mathematical computations as a function of both the pixel luminance IM.sub.HI and the low resolution luminance correction value .delta..sub.LO. Each pixel of the original image is adjusted by the corresponding high resolution luminance correction value .delta..sub.HI and a luminance adjusted image is produced.

Method and apparatus for pixel level luminance adjustment

Full color resolution of an image is realized by reconstructing an array of image data points from an image signal representing the image into a shifted array of pseudo pixels. The particular grid type of the array will dictate whether color recovery uses only vertical resolution, only horizontal resolution, vertical resolution followed by horizontal resolution, or horizontal resolution followed by vertical resolution. For each color channel, interpolation of missing color components is accomplished by first, taking a DCT of the image data points, then taking a modified IDCT of the DCT coefficients.

System and method for color recovery using discrete cosine transforms

A method and apparatus for applying an image filter to an image signal where image data terms, corresponding to the image signal, are converted by means of an overlapping operation and a scaled forward orthogonal transformation to form frequency coefficient matrices, the image filter is converted by means of a descaled orthogonal transformation to form a descaled frequency filter matrix, and the frequency coefficient matrices are multiplied by the descaled frequency filter matrix to form filtered coefficient matrices for conversion into a filtered image signal by means of an inverse orthogonal transformation process.

Method and apparatus for fast two-dimensional cosine transform filtering

The invention relates to a novel process and system for removing noise from an image by first noise modeling an image signal source to generate noise masks and LUT values characteristic of noise at different frequency levels for each channel, and then applying the stored noise masks and LUT values to an image signal for noise removal. The image is first captured as an electronic image signal by the image signal source, then represented by a pyramid structure whereby each successive level of the pyramid is constructed from DC values of the previous level, and each level of the pyramid corresponds to a different frequency band of the image signal. A Wiener variant filter using DCT transforms is used to filter DCT coefficients at each level. The image is restored with reduced noise by replacing DC values with next level IDCT coefficients then performing an IDCT on the results.

Growing community of inventors

Malden, MA, United States of America

Ibrahim Hajjahmad