Chair: Edward J. Delp, Purdue University, USA
Neal R. Harvey, University of Strathclyde (U.K.)
Stephen Marshall, University of Strathclyde (U.K.)
Corruption or noise is a common problem in any field which makes use of images. If archive film or video material is to be broadcast it is desirable to remove unwanted noise/ corruption from the original material, prior to broadcast. This paper describes a method for restoration of video images corrupted by "scratching", using a combination of non-linear techniques.
Saif S. Zahir, The University of British Columbia (Canada)
Morton Kanefsky, University of Pittsburgh (U.S.A.)
A new adaptive near-information preserving image encoder that employs optimum predictive technique and that is insensitive to the type of image or the segmentation method employed, is presented. The encoder uses a non-symmetric half plan NSHP region of support (ROS) as well as a binary image that identifies the various regions of the segmented image as either stationary (i.e., homogenous) regions or non-stationary (i.e., transition) regions. Encoding is implemented via linear predictors whose coefficients, region of support, and prediction error quantization adapt depending on pixels location in the binary image. Reconstructed images are compared with those of segmentation based two-source coding algorithms and found to be objectively and subjectively significantly better.
Bo Shen, Wayne State University (U.S.A.)
Dongge Li, Wayne State University (U.S.A.)
Ishwar K. Sethi, Wayne State University (U.S.A.)
This paper presents a video cut detection algorithm using multi-level Hausdorff distance histograms. Hausdorff distance is obtained by comparing edge points of successive frames, wherein the edge information is extracted from compressed frames directly. The use of Hausdorff distance histogram instead of the comparison of entering/exiting edge pixel counts [71 makes the algorithm more robust to complicated camera shots. The experimental results show that this algorithm can robustly tolerate rapid changes in scene brightness as well as multiple object and camera motions.
I. Khandogin, University of Wuppertal (Germany)
A. Kummert, University of Wuppertal (Germany)
D. Maiwald, STN ATLAS Elektronik GmbH (Germany)
Nonlinear image processing algorithms axe important tools for solving sophisticated problems in the domain of image analysis. In particular, they are a useful and powerful alternative to linear filtering if objects have to be detected in real time on the basis of geometrical properties. In the following part , a nonlinear image processing system designed for the automatic inspection of rail surfaces by means of on-line analysis of video sequences is presented. The inspection of the fixing devices and other components of the track are not considered here. In our application, nonlinear operations are used for image enhancement, normalization, detection of damaged areas on the rail head and filtering of time series. This paper addresses practical aspects and advantages of equalizing image histograms, morphological filtering, binarization of images, and median filtering.
Yoshinori Izumi, NHK (Japan Broadcasting Corporation) (Japan)
Masahide Naemura, NHK (Japan Broadcasting Corporation) (Japan)
Atsushi Fukuda, NHK (Japan Broadcasting Corporation) (Japan)
Yoshinobu Mizutani, NHK (Japan Broadcasting Corporation) (Japan)
Yuichi Ninomiya, NHK (Japan Broadcasting Corporation) (Japan)
Our developed noise reducer, called a dc-shift noise reducer, reduces noise by shifting the local signal level to minimize the noise level. A combination with a dc-shift technique and a median filter reduced the noise in a moving area effectively. We have successfully developed the shift noise reducer for HDTV receivers to reduce the transmission noise. Detection of the noise level in a real moving picture is a problem that must be solved to utilize the dc-shift noise reducer for real moving TV pictures. However, we have developed the noise level detection method using a statistical distribution analysis. We also introduced the detection of global and local noise levels, which contribute to the dynamic threshold E: e-filter instead of the median filter originally used. These newly developed techniques of statistical analysis for noise level detection and the dynamic threshold c -filter improved the dc-shift noise reducer for the use of real moving TV sequences.
Daben Liu, BBN Systems and Technologies (U.S.A.)
Sos Agaian, University of Texas at San Antonio (U.S.A.)
Joseph P. Noonan, Tufts University (U.S.A.)
A technique to remove block artifacts in compressed images is presented. This technique is based on a local representation of the image using an orthogonal edge basis. Small image regions are classified into several prototypes according to the coefficients of the representation. Three algorithms are proposed. One technique reduces block effects by adjusting the coefficients. The other two techniques classify the local image into several prototypes and introduces different edge-preserving filtering strategies for different types. Experiments have shown promising results with improved visual qualities and higher peak-signal-to-noise-ratio (PSNR).