Image analysis – Image transformation or preprocessing – General purpose image processor
Reexamination Certificate
2000-06-02
2004-03-23
Mehta, Bhavesh M. (Department: 2625)
Image analysis
Image transformation or preprocessing
General purpose image processor
C382S190000, C382S260000
Reexamination Certificate
active
06711306
ABSTRACT:
FIELD OF THE INVENTION
The invention relates generally to the field of image processing, and in particular to image processing techniques used for locating specific features in images.
BACKGROUND OF THE INVENTION
Extremely bright image regions can be the source of significant neutral channel errors in a scene balance algorithm (SBA). Indoor pictures photographed with a window through which bright daylight is visible are a common instance of this problem. This situation can result in a severe backlighting condition. Being able to detect these features could enable improved SBA performance as well as provide other useful information about the scene, such as the image location (specifically, indoor vs. outdoor), the orientation of the image, and the detection of the main subject.
It is known to use the brightness of particular portions of the scene in order to adjust exposure. For instance, in U.S. Pat. No. 5,227,837, the brightness of an object is estimated by using the brightness information recorded at the time of taking the object image. As a result, it is possible to discriminate correctly those measurement points which are unnecessary for scene discrimination or exposure control, thereby realizing highly precise exposure control. In doing this, the object is subjected to photometry at divisional areas at the time of taking the object image, and the obtained photometric values at divisional areas are used in discriminating the type of recorded scene.
A number of different techniques are well known and available for the detection and location of objects in a scene. For example, the use of edge detection combined with neural networks is disclosed in U.S. Pat. No. 5,481,628 for object detection. The use of vertex identification for object detection is taught by U.S. Pat. No. 5,838,830. U.S. Pat. No. 5,848,190 relies on pattern matching for object detection, while U.S. Pat. No. 5,877,809 extracts a target object based on a calculated object distance. While generally addressing the subject of object detection, none of these references address the unique issues of window detection.
Since there is a rich literature on object detection in the field of computer vision, it is tempting to think of window detection in those terms, with “windows” as being just another type of object whose detection can be performed by applying traditional methods. Unfortunately, window detection is qualitatively different from other types of object recognition in that windows can come in a seemingly infinite variety of appearances. WordNet (a public domain online hyperlinked dictionary) defines 22 different hyponyms (kinds of) for windows. The main feature of a window in an image isn't even the window itself, since that is by definition transparent, or largely so. In fact, Webster's dictionary defines a window as “an opening in the wall of a building”. Interestingly, WordNet takes the opposite tack and defines a window as a “framework of wood or metal that contains a glass windowpane”.
It would appear that windows are recognized by humans through some high level semantic processing, largely by their edges and the discontinuity between the image content inside and outside of the window region. However, recognizing windows simply as panes of glass is not practical since glass is normally transparent. Nor can windows be recognized solely by their content since that can be arbitrary. And unfortunately, it is the rule rather then the exception for window edges to be obscured, whether by drapes, curtains, shades and the like, or by occluding objects such as plants or lamps, or even by the main subject of the picture. As the window itself can look out onto anything, there are really very few low-level clues as to its presence. Worse still, many of the low-level features that can be used to characterize windows, such as brightness, corners, or vertical edges, also characterize many other common image features. All of these issues combine to make general “window detection” a very difficult problem indeed.
SUMMARY OF THE INVENTION
It is an object of the invention to automatically identify bright windows in photographic images.
It is a further object to segment a photographic image into regions followed by an analysis of the morphological characteristics of the regions in order to identify a bright window.
The present invention is directed to overcoming one or more of the problems set forth above. Briefly summarized, according to one aspect of the present invention, a method for automatically processing a digital image to locate one or more windows that are substantially brighter than their surroundings begins by processing the digital image to compute a feature image identifying the location of features in the image based on a weighted contribution of edge information, brightness information, information corresponding to spatial activity and occlusion boundary information. Then the feature image is processed with one or more morphological filtering operations to provide edge smoothing and noise removal, thereby generating a filtered image. The filtered image is processed to identify zero or more regions, wherein the presence of one or more regions provides a segmented image, and the segmented image is logically combined with the occlusion boundary information to provide one or more window candidates. Actual windows are deduced by verifying one or more of the window candidates based on their mean intensity relative to the mean intensity of the digital image.
The invention provides the advantage of automatically determining in a photographic image the presence and location of windows that are substantially brighter than their surroundings. Furthermore, it detects the presence of bright windows without the use to template matching or any a priori knowledge of other objects in the scene.
REFERENCES:
patent: 4373810 (1983-02-01), Shreve
patent: 5227837 (1993-07-01), Terashita
patent: 5305204 (1994-04-01), Ohhashi
patent: 5392091 (1995-02-01), Iwasaki
patent: 5426879 (1995-06-01), Hecker
patent: 5475812 (1995-12-01), Corona et al.
patent: 5481628 (1996-01-01), Ghaderi
patent: 5715377 (1998-02-01), Fukushima et al.
patent: 5828793 (1998-10-01), Mann
patent: 5838830 (1998-11-01), Qian et al.
patent: 5877809 (1999-03-01), Omata et al.
patent: 5974190 (1999-10-01), Maeda et al.
patent: 5978443 (1999-11-01), Patel
patent: 6147705 (2000-11-01), Krauter et al.
patent: 6240217 (2001-05-01), Ercan et al.
patent: 6297889 (2001-10-01), Loce et al.
Eastman Kodak Company
Mehta Bhavesh M.
Patel Kanji
Woods David M.
LandOfFree
Automatic bright window detection does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Automatic bright window detection, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Automatic bright window detection will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3207096