Abstract: A method and apparatus is described that categorizes images by extracting regions and describing the regions with a set of 15-dimensional image patch feature vectors, which are concatenations of color and texture feature vectors. By comparing the image patch feature vectors in images with similarly-obtained image patch vectors in a Gaussian mixture based model pool (obtained in an image patch modeling phase), the images may be categorized (in an image patch recognition phase) with probabilities relating to each image patch. Higher probabilities are likelier correlations. The device may be a single or multiple core CPU, or parallelized vector processor for characterizing many images. The images may be photographs, videos, or video stills, without restriction. When used real-time, the method may be used for visual searching or sorting.

Abstract: A method to estimate segmented motion uses phase correlation to identify local motion candidates and a region-growing algorithm to group small picture units into few distinct regions, each of which has its own motion according to optimal matching and grouping criteria. Phase correlation and region growing are combined which allows sharing of information. Using phase correlation to identify a small number of motion candidates allows the space of possible motions to be narrowed. The region growing uses efficient management of lists of matching criteria to avoid repetitively evaluating matching criteria.

Abstract: A system and method for effectively performing an image identification procedure includes an image matching manager that derives source characteristics for a source image and target characteristics for target images. The image matching manager compares the source characteristics and the target characteristics to determine whether the source image matches any of the target images. The source characteristics and the target characteristics may include color-space characteristics and curve-space characteristics from the respective images. A processor of an electronic device typically controls the image matching manager to effectively perform the image identification procedure.

Abstract: Efficient representation of color digital pathology images (DPI) is described herein, which is accomplished by exploiting properties unique to such images. The method decomposes the data into constituent parts whose relative importance is able to be specified, allowing the data to be accurately represented with less bit precision, less spatial resolution or less spectral resolution. Two specific areas where the method is able to be utilized include: (1) more-efficient image compression; and (2) more efficient processing of the data. Efficient image compression is accomplished by assigning fewer bits to less-important colors. Efficient data processing is accomplished by processing only those colors, or combinations of colors, that are deemed important.

Abstract: A method of and an apparatus for automatically estimating the tilting angle in tilted images enables level correction of the images. A preferred orientation of objects in an image and the deviation of the current orientation from the preferred orientation is determined by tilt image analysis without object recognition. Tilt image analysis includes several steps such as gradient feature computation, line segment tracking, line segment estimation and orientation deviation estimation. Once the tilt angle is determined, the image can be corrected so that an object or scene is not tilted or is tilted by only the appropriate amount.

Abstract: A system and method for effectively performing a scene rectification procedure comprises an image manager that includes a segmentation module, a label module, and a rectification module. The segmentation module initially performs a segmentation procedure upon an image to produce corresponding sub-scenes. The label module then categorizes the sub-scenes by assigning initial labels without utilizing context information from other sub-scenes in the image. The rectification module performs a semantic grouping procedure upon the sub-scenes to produce semantic group nodes corresponding to pairs of the sub-scenes that have a predefined semantic relationship. The rectification module converts a sub-scene graph of the sub-scenes into a semantic graph that includes the semantic group nodes. The rectification module then performs a rectification procedure to convert the initial labels of the sub-scenes into rectified labels.

Abstract: Low complexity edge detection and DCT type selection method to improve the visual quality of H.264/AVC encoded video sequence is described. Encoding-generated information is reused to detect an edge macroblock. Variance and Mean Absolute Difference (MAD) of one macroblock shows a certain relationship that is able to be used to differentiate the edge macroblock and the non-edge macroblock. Also, the variance difference of neighbor macroblocks provides a hint for edge existence. Then, a block-based edge detection method uses this information. To determine the DCT type for each block, the detected edges are differentiated as visual obvious edge, texture-like edge, soft edge and strong edge. 8×8 DCT is used for texture-like edges and the 4×4 DCT is used for all the other edges. The result is an efficient and accurate edge detection and transform selection method.

Abstract: A “Bokeh-Aji” image is one in which the region of interest is in focus and the background is out of focus. Detection of “Bokeh-Aji” type images and then isolation to the region of interest area in a low complexity way without any human intervention is beneficial. A set of tools for performing this task include SAD and high pass filtering based in-focus/out-of-focus area separation, in-focus/out-of-focus block distribution based “Bokeh-Aji” shot detection and region of interest isolation. By effectively integrating these tools together, the “Bokeh-Aji” images are successfully identified, and the region of interest area is successfully isolated.

Abstract: Apparatus and methods for generating a shutter-time compensated high spatial resolution (HR) image output by enhancing lower spatial resolution (LR) video images with information obtained from higher spatial resolution still images which are temporally decimated. Super-resolved images and LR SAD information is generated from the LR images and used for directing the extracting of information from the temporally decimated HR images to enhance the spatial resolution of the LR images in a blending process. By way of example blending can comprise: motion estimation, motion compensation of a temporally displaced HR still images and a super-resolved (SR) image input, transformation (e.g., DCT), generating motion error output, blending motion compensated images in response to LR motion error information; inverse-transformation into a shutter-time compensated HR video image output. Accordingly, a more cost effective solution is taught for obtaining a desired shutter time and video resolution.

Abstract: An apparatus and method for detecting “Object Portraits” (photographs or images with a stand-out object of interest or a set of stand-out objects of interest) is described. A set of tools has been developed for object of interest detection, including “Sunset-like” scene detection, pseudo-color saturation-based detection and object of interest isolation, block intensity based detection and object of interest isolation. By effectively integrating these tools together, the “Object Portrait” images and “Non-Object Portrait” images are successfully identified. Meaningful object of interest areas are thereby successfully isolated in a low complexity manner without human intervention.

Abstract: Efficient representation of color digital pathology images (DPI) is described herein, which is accomplished by exploiting properties unique to such images. The method decomposes the data into constituent parts whose relative importance is able to be specified, allowing the data to be accurately represented with less bit precision, less spatial resolution or less spectral resolution. Two specific areas where the method is able to be utilized include: (1) more-efficient image compression; and (2) more efficient processing of the data. Efficient image compression is accomplished by assigning fewer bits to less-important colors. Efficient data processing is accomplished by processing only those colors, or combinations of colors, that are deemed important.

Abstract: An image alignment method includes computationally efficient methods of achieving high-accuracy local motion estimates by using phase correlation. The method also estimates motion reliability that allows a generic robust model fitting algorithm to produce more accurate results while operating much more efficiently. One of three methods are used to determine sub-pel motion estimation with improved accuracy. Each of the sub-pel motion estimation methods uses phase correlation, and are based on fitting computationally efficient 2-D quadratic surfaces to a phase correlation surface. A pre-filter is applied which shapes the phase correlation surface to enable appropriate fitting to the quadratic surface. Bias is also compensated for prior to applying a sub-pel motion estimation method. The method also estimates the reliability of the sub-pel motion estimates determined using phase correlation.

Abstract: A system and method for effectively performing an integrated segmentation procedure comprises an image segmenter that includes a texture modeler, a contrast modeler, and a model integrator. The texture modeler creates a texture model based upon an original image. Similarly, the contrast modeler creates a contrast model based upon the original image. The model integrator then performs a model integration procedure to create a final segmented image by integrating the texture model and the contrast model according to a calculated texture model metric. A processor of an electronic device typically controls the image segmenter to perform the integrated segmentation procedure.

Abstract: A system and method for effectively performing an image identification procedure includes an image matching manager that derives source characteristics for a source image and target characteristics for target images. The image matching manager compares the source characteristics and the target characteristics to determine whether the source image matches any of the target images. The source characteristics and the target characteristics may include color-space characteristics and curve-space characteristics from the respective images. A processor of an electronic device typically controls the image matching manager to effectively perform the image identification procedure.

Abstract: A system and method for effectively implementing a stroboscopic visual effect with a television device includes a strobe engine that analyzes video data to create a sequence of stroboscopic images based upon motion information from the video data. The television utilizes a display manager to present the stroboscopic images and the video data on a display device during a strobe display mode. A processor device of the television typically controls the operations of the strobe engine and the display manager to implement the stroboscopic visual effect.

Abstract: An adaptive picture mode selection transcoder selects an encoding mode in a second format for frames of video previously encoded in a first format by determining a magnitude of interlacing phenomenon in the using picture information obtained during decoding of the video from the first format. In one aspect, the picture information includes discrete cosine coefficients for macroblocks in the frame. In another aspect, the picture information includes an encoding mode for the macroblocks in the first format. In yet another aspect, the picture information includes motion vector information for the macroblocks. In still another aspect, the determining is specific to an encoding mode for the frame.

Abstract: A video system includes: analyzing video data, having a block; performing a transition change detection for determining a spatial intensity transition within the block; performing a block-wise similarity measurement on the block in the video data for identifying a blocking artifact; and filtering with a two dimensional cross filter every pixel in the block for removing the blocking artifact.

Abstract: A system and method for effectively performing a scene rectification procedure comprises an image manager that includes a segmentation module, a label module, and a rectification module. The segmentation module initially performs a segmentation procedure upon an image to produce corresponding sub-scenes. The label module then categorizes the sub-scenes by assigning initial labels without utilizing context information from other sub-scenes in the image. The rectification module performs a semantic grouping procedure upon the sub-scenes to produce semantic group nodes corresponding to pairs of the sub-scenes that have a predefined semantic relationship. The rectification module converts a sub-scene graph of the sub-scenes into a semantic graph that includes the semantic group nodes. The rectification module then performs a rectification procedure to convert the initial labels of the sub-scenes into rectified labels.

Abstract: A method and apparatus for generating a super-resolution image are provided. The method may include obtaining a first set of RAW data representing a first image captured at a first resolution and obtaining, from the first set of RAW data, at least one first sample of data associated with the first image. The method may also include obtaining a second set of RAW data representing a second image captured at the first resolution, and performing image registration as a function of the first set of RAW data and the second set of RAW data so as to obtain at least one second sample of data associated with the second image. The first set of RAW data is used as a reference for the second set of RAW data. The method further includes combining the at least one first sample of data with at least one second sample of data to form a collection of samples, and interpolating the collection of samples to form the super-resolution image.

Abstract: Implementing color effects in compressed digital video is improved upon by re-using the original video's compression parameters during the re-encoding stage, such that the parameters do not need to be re-estimated by the encoder. This improved method reduces complexity and also improves quality. Quality is improved due to re-use of the compression parameters since accumulated error which is common when re-encoding compressed video is prevented. For digital negatives, the effect is able to be implemented even more efficiently.