An algorithm for covert digital audio watermarking is presented. The method embeds a low-rate (in the neighborhood of 20–30 b/sec) watermark via perceptually insignificant long term multiband phase modulation, The watermark is very difficult to either discover or recover without the original (unmodulated) signal. The algorithm is shown to be robust to perceptual coding.

A spatial noise shaping (SNS) method based on human visual sensitivity is presented. The method exploits the capability of frequency domain linear prediction for spatial envelope retrieval. It effectively shapes (or hides) the noise of an image in areas which are not sensitive to human vision so that the resulting image is more pleasant to human eyes. The noise comes from the processing of the image, and it can be either separable like the additive noise pattern in image watermarking or nonseparable like the quantization noise in image coding. An application of the algorithm is demonstrated in the paper by using it to enhance image coders. Images decoded from the SNS incorporated coders have superior perceived quality than those without using SNS.

We have developed and demonstrated that spatial noise shaping (SNS) can enhance the performance of an image coder. SNS runs open-loop 2-D linear prediction (LP) in the frequency domain instead of in the time domain as compared to the standard LP used in speech and image coding. This predictive analysis/synthesis process over frequency possesses two important properties which result in a decoded image with superior quality compared to the one without using SNS. The first property is that since SNS can preserve the spatial structure of the image in the filter, those desirable details and sharp edges won't be damaged by the image coder and will be restored to the decoded image at the synthesis stage of SNS. The other unique property of SNS is that it effectively adapts the spatial structure of the quantization noise to that of the image roughness. Therefore, a more efficient use of masking effects can be exploited. In this paper we first explain how to use the noise shaping capability of SNS to hide the quantization noise in visually insensitive areas, such as textures and edges, but not in flat areas. Then, the techniques used to stabilize the 2-D LP filter are discussed. Finally, we show that SNS is applicable to either spectral domain coders or time domain coders, such as wavelets.

A spatial noise shaping (SNS) method based on human visual sensitivity is presented. It effectively shapes (or say hides) the noise of an image in areas which are not sensitive to human visual perception so that the resulting image is more pleasant to human eyes. The noise comes from the processing of the image, and it can be either separable like the additive noise pattern in image watermarking or non-separable like the quantization noise in image coding. An application of the algorithm is also demonstrated by using it to enhance image coders. Images decoded from the SNS incorporated coders possess a superior quality to those which do not use SNS.

We have studied and concluded that the time domain cross channel prediction is generally not applicable to perceptual audio coding. From the statistical analysis, correlation among certain channels seems to be enough to provide some coding gain and the prediction also successfully reduces the total energy of the signal to be coded. But this energy decrease and its resulting bit reduction mainly happens in the low frequency bands. In fact, the whitening effect of the prediction filter increases either the energy or the energy peaks in the high frequency region. As a result, more bits are required to code the high-frequency part of the signal and this increase outpaces the bit reduction realized in the low frequency region. Therefore, using cross channel linear prediction in the multichannel perceptual audio coding does not provide a net coding gain.

In this work, discriminative training is studied to improve the performance of our pseudo two-dimensional (2-D) hidden Markov model (PHMM) based text recognition system. The aim of this discriminative training is to adjust model parameters to directly minimize the classification error rate. Experimental results have shown great reduction in recognition error rate even for PHMMs already well-trained using conventional maximum likelihood (ML) approaches.

Two efficient algorithms are presented for enhancing color images and gray-level scanned document images. For color images, a generic approach for enhancing both contrast and color saturation of color images is first developed. For real time application in NTSC devices, an efficient algorithm is then proposed to optimally enhance a color image in its native YCbCr color space without any color space conversion. As to enhance a gray-level text image, an algorithm is also proposed to successfully remove the blurriness, and maximize the contrast in real time. The resulting image possesses a much higher readability than its original form. Both algorithms achieve highly satisfactory results without introducing excess complexity. The simplicity of the algorithm and its ability to produce better results than the conventional approaches make it an excellent candidate for practical implementation in real-time applications. Experimental results are given to demonstrate the superiority of the proposed algorithms.

The pseudo two dimensional hidden Markov model (PHMM) is extended to directly recognize poorly-printed gray-scale document images. The N-best hypotheses search, coupled with duration correction, is also developed to find best candidates. Experimental results have demonstrated that the performance of the new system has been significantly improved when compared to the original PHMM system [Kuo and Agazzi, 1994] using binary images as inputs. The recognition rate improves from 97.7% to 100%, over a testing set with similar blur and noise conditions as the training set. For a testing range far outside the training one, it improves from 89.59% to 98.51%, which also demonstrates the robustness of the proposed system.

An algorithm for robust machine recognition of keywords embedded in a poorly printed document is presented. For each keyword, two statistical models, called pseudo 2-D hidden Markov models, are created for representing the actual keyword and all the other extraneous words, respectively. Dynamic programming is then used for matching an unknown input word with the two models and for making a maximum likelihood decision. Although the models are pseudo 2-D in the sense that they are not fully connected 2-D networks, they are shown to be general enough in characterizing printed words efficiently. These models facilitate a nice "elastic matching" property in both horizontal and vertical directions, which makes the recognizer not only independent of size and slant but also tolerant of highly deformed and noisy words. The system is evaluated on a synthetically created database that contains about 26000 words. Currently, the authors achieve a recognition accuracy of 99% when words in testing and training sets are of the same font size, and 96% when they are in different sizes. In the latter case, the conventional 1-D HMM achieves only a 70% accuracy rate.>

An algorithm for robust machine recognition of keywords embedded in a poorly printed document is presented. For each keyword, two statistical models, named hidden Markov models (HMMs), are created for representing the actual keyword and all the other extraneous words, respectively. Dynamic programming is then used for matching an unknown input word with the two models and making a maximum likelihood decision. Both the 1D and pseudo-2D HMM approaches are proposed and tested. The 2D models are shown to be general enough in characterizing printed words efficiently. These pseudo-2D HMMs facilitate an elastic matching property in both the horizontal and vertical directions, which makes the recognizer not only independent of size and slant but also tolerant of highly deformed and noisy words. The system is evaluated on a synthetically created database. Recognition accuracy of 99% is achieved when words in testing and training sets are in the same font size, and 96% is achieved when they are in different sizes. In the latter case, the 1D HMM achieves only a 70% accuracy rate.>

The authors introduce a method to render optical character recognition algorithms based on pseudo two-dimensional hidden Markov models (PHMMs) independent of image transformations such as scaling, translations, slant, vertical and/or horizontal stretching, etc. Estimation of transformation parameters and image normalization are performed simultaneously with recognition. When combined with a previous method for joint segmentation and recognition of connected and degraded text, this method can be used to recognize extremely degraded documents that include characters affected by various geometric transformations. Experiments with isolated characters where scaling, slant angle, and translation are varied over ranges of 4: 1, 0/spl deg/ to 45/spl deg/, and 0 to 40 pixels respectively, are presented. Also presented are experiments with connected text where images have been affected both by geometric and stochastic distortions of various degrees, that show the high effectiveness of this technique.>

The authors apply a hidden Markov model (HMM) and a level-building dynamic programming algorithm to the problem of robust machine recognition of connected and degraded characters forming words in a poorly printed text. A structural analysis algorithm is used to segment a word into sub-character segments irrespective of the character boundaries, and to identify the primitive features in each segment such as strokes and arcs. The states of the HMM for each character are statistically represented by the sub-character segments and the state characteristics are obtained by determining the state probability functions based on the training samples. A level-building dynamic programming algorithm combines word-segmentation and recognition in one operation and chooses the best probable grouping of characters for recognition of an unknown word. The computer experiments demonstrate the robustness and effectiveness of the system for recognizing words formed by degraded and connected characters.>

A new iterative algorithm for image restoration and point spread function (PSF) estimation is presented. The method initially restores the image with the expectation maximization (EM) algorithm. The resulting estimate is then significantly improved by using the adaptive row action projection (RAP) algorithm which is based on the theory of projection onto convex sets (POCS). The PSF is re-estimated using a least squares technique. Computer simulations show the method's competitiveness in restoring degraded images and estimating the PSF from noisy blurred images with unknown PSF.>