Abstract: A method for performing highlight restoration on a digital image includes comparing the pixels in the image with a saturation level value to identify saturated pixels. A saturation map of saturated pixels is generated. Each selected saturated pixel is identified as a restorable pixel only if at least one color channel of the pixel is unsaturated. For each restorable pixel, a group of the closest unsaturated pixels above, below, to the left, and to the right of the select saturated pixel is identified. A replacement pixel value is generated for each saturated color channel of the restorable pixel, using a combination of the pixel values of the unsaturated color channels of the restorable pixel and the pixel values of the corresponding color channels of the nearby unsaturated pixels.

Abstract: A method for suppressing structured noise in a digital image includes creating a smoothed version of the original image. Monotonic and slowly-varying image regions are detected by analyzing a residual image which is the function of the original image and its smoothed version. A local window is defined in each pixel location identified in the thresholding process as the location with structured noise and samples inside the window are randomly permuted to randomize the noise structures. A noise-filtered version of the original residual image is generated. The noise-filtered residual and the smoothed version of the original image are combined to produce a final image.

Abstract: Analyzing an input image, the input image being one of a digitized image stored in a memory or a scanned image from a scanner. Forming a feature image from the input image by dividing the input image into a plurality of blocks of pixels, thus associating each block of pixels in the input image with a single pixel in the feature image, and outputting the feature image for further analysis or storage in a memory. Example embodiments extract and analyze features from a document image to detect particular characteristics associated with the page area, the distortion area, and the book spine area. Extracted features can be further analyzed to detect document characteristics at the paragraph, line, word, and character levels.

Abstract: Automatic red-eye object classification in digital photographic images. A method for classifying a candidate red-eye object in a digital photographic image includes several acts. First, a candidate red-eye object in a digital photographic image is selected. Next, RGB pixels of the candidate red-eye object are converted into YUV pixels. Then, the YUV pixels satisfying a constraint that is a function of the YUV pixels are summed. Next, the sum is determined to be greater than or equal to a scaled version of the total number of YUV pixels in the candidate red-eye object. Finally, the candidate red-eye object is transformed into a true red-eye object.

Abstract: Disclosed are methods, devices, and computer program products for red-eye detection in a digital image. In one example embodiment, a method for detecting a red-eye effect in a digital image includes several acts. First, red pixels having a predetermined degree of redness are identified in the image. Next, redness contrast is detected with respect to each of the red pixels and redness is then enhanced for those red pixels having a predetermined level of redness contrast. The pixels identified as being red are then further refined by applying another redness threshold based on one or more color characteristics associated with the red pixels. The refined set of red pixels may then be partitioned into a set of one or more candidate red-eye objects. A candidate red-eye object may be removed as a false positive based on geometric constraints associated with red-eye objects and/or proximity of the object to pixels with human skin-like color tones.

Abstract: Disclosed are methods, devices, and computer program products for red-eye detection in an image. In one example embodiment, a method for detecting red-eye objects in an image includes several acts. First, a set of candidate red-eye objects identified in the image is received. Then, features are extracted from the candidate red-eye objects and, with a plurality of classifiers, a false red-eye object is eliminated from the set of candidate red-eye objects based on the extracted features. First and second ones of the plurality of classifiers are optimized for classifying objects in a first range of sizes using first and second ones of the extracted features, respectively. Furthermore, third and fourth ones of the plurality of classifiers are also optimized for classifying objects using the first and second ones of the extracted features, respectively, but for objects in a second range of sizes.

Abstract: Analyzing an input image, the input image being one of a digitized image stored in a memory or a scanned image from a scanner. Forming a feature image from the input image by dividing the input image into a plurality of blocks of pixels, thus associating each block of pixels in the input image with a single pixel in the feature image, and outputting the feature image for further analysis or storage in a memory. Example embodiments extract and analyze features from a document image to detect particular characteristics associated with the page area, the distortion area, and the book spine area. Extracted features can be further analyzed to detect document characteristics at the paragraph, line, word, and character levels.

Abstract: A method for performing fast detail-preserving filtering of an input digital image includes, for each pixel in the image, calculating the difference between the selected pixel and each of its four neighboring pixels located above, left, right, and below the selected pixel, calculating a scaled weighted sum of differences between the actual pixel and its four neighbors, where for each neighboring pixel the weight is an edge-sensing function having a data-adaptive scaling parameter function, and adding the weighted sum of differences to the value of the selected pixel.

Abstract: Automatic red-eye object classification in digital images using a boosting-based framework. In a first example embodiment, a method for classifying a candidate red-eye object in a digital photographic image includes several acts. First, a candidate red-eye object in a digital photographic image is selected. Next, a search scale set and a search region for the candidate red-eye object where an eye object may reside is determined. Then, the number of subwindows that satisfy an AdaBoost classifier is determined. This number is denoted as a vote. Next, the maximum size of the subwindows that satisfy the AdaBoost classifier is determined. Then, a normalized threshold is calculated by multiplying a predetermined constant threshold by the calculated maximum size. Next, the vote is compared with the normalized threshold. Finally, the candidate red-eye object is transformed into a true red-eye object if the vote is greater than the normalized threshold.

Abstract: A method for replacing defective pixels in a digital color image includes determining whether each pixel has defective data in a selected color channel; for the pixel, determining whether a first reference color channel exists and, if so, correcting the defective data by defining a group of neighboring pixels; for each of m neighboring pixels having non-defective data in the selected color channel and the reference color channel, computing a sum of the differences between the non-defective data in the selected color channel and the non-defective data in the first reference color channel; adding the sum of the differences divided by m to the non-defective data value from the first reference color channel to obtain a result; dividing the result by two to obtain a substitution data value; and substituting the substitution data value for the defective data.

Abstract: An image processing method that demosaicks a mosaic input image to generate a full color output image. The image processing method calculates both vertical and horizontal luminance-chrominance difference components for each pixel of the mosaic input image. Next, the image processing method calculates an enhanced version of both vertical and horizontal luminance-chrominance difference components for each pixel of the mosaic input image. Then, the image processing method interpolates a G component for each of the original R and B components. Next, the image processing method detects a signal overshoot or undershoot in each interpolated G component and to clamps each interpolated G component with a detected signal overshoot or undershoot to the closest neighboring original G component. Next, the image processing method interpolates missing R and/or B components in each pixel location of the captured image.

Abstract: An image processing apparatus that receives mosaic image data having settings of only one color component, R, G, or B, in each pixel and subjects the received mosaic image data to a series of image processing to generate color image data with settings of all the three color components, R, G, and B, in each pixel; wherein the mosaic image data has the form of a Bayer color filter array; and the image processing apparatus includes: a vertical-direction color difference component computation module; a horizontal-direction color difference component computation module; an edge direction determination module; a color component interpolation module; an oblique edge pixel detection module; an oblique edge direction determination module; and an oblique edge pixel interpolation correction module.

Abstract: An image processing method that demosaicks a mosaic input image of R, G, and B components to generate a full color output image. The image processing method calculates both vertical and horizontal luminance-chrominance difference components for each pixel of the mosaic input image. Next, the image processing method calculates an enhanced version of both the vertical and horizontal luminance-chrominance difference components for each pixel of the mosaic input image. Next, the image processing method evaluates the variations in the enhanced luminance-chrominance difference components in order to create an edge directional map indicating the direction in which demosaicking should be performed. Then, the image processing method interpolates a G component for each of the pixels with the original R and B components using the edge directional map.

Abstract: Enlarging a digital image. In one example embodiment, a method for enlarging a digital image includes various acts. First, an enlargement factor ? is selected for an input image. Next, a pixel in the input image is selected. Then, a supporting window is placed over the input image. Next, a ?×? block of output pixels is produced. Each pixel in the block of output pixels is produced using a set of ?2 distinct weight matrices. Then, the block of output pixels is assembled into an output image. Next, the acts of selecting a pixel, placing the supporting window, producing a block of output pixels, and assembling the block of output pixels into the output image are repeated for each of the remaining pixels in the input image.

Abstract: The image processing procedure of the invention receives mosaic image data and calculates a vertical-direction color difference component with regard to each of pixel columns in the mosaic image data in a vertical direction and a horizontal-direction color difference component with regard to each of pixel rows in the mosaic image data in a horizontal direction.

Abstract: Joint automatic demosaicking and white balancing. In one example embodiment, a digital image processing method includes several acts. First, directional color correlation signals, global gains, and orientations of edges are calculated in a CFA input image. Next, missing luminance components in CFA chrominance locations are demosaicked along edges in the input image using CFA chrominance components and the directional color correlation signals. Then, the CFA chrominance components are white-balanced using the demosaicked luminance components, the CFA chrominance components, and white-balancing gains expressed as a function of the global gains and local gains calculated directly from a pixel under consideration. Next, missing chrominance components in CFA chrominance locations in the input image are demosaicked. Finally, missing chrominance components in CFA luminance locations in the input image are demosaicked.

Abstract: An image processing procedure receives mosaic image data and calculates vertical and horizontal-direction color difference components for each pixel. The image processing procedure subsequently selects an R pixel or a B pixel from the mosaic image data, and compares a variation of the vertical-direction color difference component with a variation of the horizontal-direction color difference component with regard to each of at least the selected pixels to detect edge directions of the at least selected pixels. The edge directions thus obtained are collected in an edge direction map, and then the edge directions are compared with the surrounding edge directions to remove edge noise in advance. The image processing procedure refers to the detected edge directions, and interpolates a missing color component in each pixel of the mosaic image data with the settings of one color component in each pixel in the mosaic image data.

Abstract: The image processing procedure of the invention receives mosaic image data and calculates a vertical-direction color difference component with regard to each of pixel columns in the mosaic image data in a vertical direction and a horizontal-direction color difference component with regard to each of pixel rows in the mosaic image data in a horizontal direction. The mosaic image data is expressed by a combination of pixel columns with alternate arrangement of pixels of a G component and pixels of an R component in the vertical direction, pixel columns with alternate arrangement of pixels of the G component and pixels of a B component in the vertical direction, pixel rows with alternate arrangement of pixels of the G component and pixels of the R component in the horizontal direction, and pixel rows with alternate arrangement of pixels of the G component and pixels of the B component in the horizontal direction.

Abstract: Automatic red-eye object classification in digital images using a boosting-based framework. In a first example embodiment, a method for classifying a candidate red-eye object in a digital photographic image includes several acts. First, a candidate red-eye object in a digital photographic image is selected. Next, a search scale set and a search region for the candidate red-eye object where an eye object may reside is determined. Then, the number of subwindows that satisfy an AdaBoost classifier is determined. This number is denoted as a vote. Next, the maximum size of the subwindows that satisfy the AdaBoost classifier is determined. Then, a normalized threshold is calculated by multiplying a predetermined constant threshold by the calculated maximum size. Next, the vote is compared with the normalized threshold. Finally, the candidate red-eye object is transformed into a true red-eye object if the vote is greater than the normalized threshold.

Abstract: Automatic red-eye object classification in digital photographic images. In a first example embodiment, a method for classifying a candidate red-eye object in a digital photographic image includes several acts. First, a candidate red-eye object in a digital photographic image is selected. Next, RGB pixels of the candidate red-eye object are converted into YUV pixels. Then, the YUV pixels satisfying a constraint that is a function of the YUV pixels are summed. Next, the sum is determined to be greater than or equal to a scaled version of the total number of YUV pixels in the candidate red-eye object. Finally, the candidate red-eye object is transformed into a true red-eye object.