Back to results
US 6,647,140 B1Granted
Spectrum inverter apparatus and method
AssigneeBank One Corporation
InventorsShen-Ge Wang, Gaurav Sharma
Life of this patent
Estimating the term…
Abstract
A spectrum inverter receives optical values from a scanning device and applies a spectrum inversion method to the optical values. The spectrum inversion method is selected based on a medium identifier that identifies the medium upon which the scanned image is formed. The resulting spectral space values are then output and may be used by an image output device to reproduce the scanned image.
Claims21
An apparatus that inverts the spectrum of image data, comprising:
- an image data input device that inputs image data of an input image in an input color space;
- a medium identification device that identifies an input image medium from which the input image data was obtained;
- and
- a spectrum inversion device that selects one of a plurality of spectrum inversion methods to convert the input image data into output image data in a spectral space based on the identification of the input image medium, the spectrum inversion device including a neural network, wherein the neural network includes input nodes, hidden nodes and output nodes, connections between the input nodes and the hidden nodes are represented by a 3×10 matrix, connections between the hidden nodes and the output nodes are represented by a 10×36 matrix, entries of the 3×10 matrix and the 10×36 matrix are determined to minimize a merit function, the merit function is a weighted root-mean-square (RMS) spectral error defined as:
- ∑ ∑ i = 1 36 w ( λ i ) * ( R desired ( λ i ) - R output ( λ i ) ) 2
- where λi, i=1, 2, . . . , 36 are sampling wavelengths, Rdesired (λi) and Routput (λi) are the spectral reflectance of desired output and actual output, and w(λi) is a weight factor.
3 more elements
A method for inverting the spectrum of image data, comprising:
- inputting image data of an input image, the image data being in an input color space;
- identifying an input image medium from which the image data was obtained;
- selecting one of a plurality of spectrum inversion methods based on the identification of the input image medium;
- and
- converting the input image data into output image data using a neural network, the output image data being in a spectral space, wherein the neural network includes input nodes, hidden nodes and output nodes, connections between the input nodes and the hidden nodes are represented by a 3×10 matrix, connections between the hidden nodes and the output nodes are represented by a 10×36 matrix, entries of the 3×10 matrix and the 10×36 matrix are determined to minimize a merit function, the merit function is a weighted root-mean-square (RMS) spectral error defined as:
- ∑ ∑ i = 1 36 w ( λ i ) * ( R desired ( λ i ) - R output ( λ i ) ) 2
- where λi, i=1, 2, . . . , 36 are sampling wavelengths, Rdesired (λi) and Routput (λi) are the spectral reflectance of desired output and actual output, and w(λi) is a weight factor.
4 more elements
Dependencies are inferred from the claim text.