Jianwei Yang

A Watermarking Scheme Based on Two-dimensional Wavelet Filter Parametrization

In this paper, a parametrization of two-dimensional wavelet filter system is used as a method to protect wavelet-based watermarks against unauthorized detection. This system is developed in terms of a novel transformation-Shift Unitary transform (SUT) of Conjugate Quadrature filter (CQF). The commonly used wavelet filters are only special cases of this system. Based on this system, a watermarking scheme is described to embed watermark into low frequency sub-bands of wavelet transformation. We overcome degradation problem by performing median filtering to the lowest frequency sub-band of wavelet transform and embed watermark into visually insensitive locations. Experiments show this method is robust to compression,median-filtering etc.

An affine invariant weighted approach for estimating parameters under affine distortions

In this paper, an affine invariant weighted method is proposed to estimate affine transformation parameters. Affine covariant images are constructed from the original image by assigning an affine invariant weight to each point in the image. Based on the first order moment of these affine covariant images, a system of linear equations is constructed to estimate the affine transformation parameters. The affine covariant images are derived in polar coordinate system, and weights of points on the same radial direction can be calculated together. In compare with the cross-weighted moment method, the proposed method yields a smaller computation burden. Experimental results also show that the proposed method can be used for the recovery of affine transformation parameters by a proper selection of parameters.

A Watermarking Scheme Based on Median Filtering

In this paper, we embed watermark into wavelet DC components. The DC area is not suitable for embedding watermark because of severe visual degradation. We overcome degradation problem by performing median filtering to the lowest frequency sub-band of wavelet transform, and embed watermark into visually insensitive locations. To embed a visually meaningful logo as watermark, we carried out Arnold transform to the logo. Experiments show that our method is robust to JPEG compression, median-filtering, and sharpening etc.

Affine Invariant Ring Fourier Descriptors

The traditional affine-invariant Fourier descriptor is contour-based. It can not be applied to objects with several components. In this paper, a region-based Affine Invariant Ring Fourier Descriptor (AIRFD) is put forward to extract affine invariant features. A set of affine invariant closed curves is constructed from the object. Prior to the extraction of features, the derived closed curves are parameterized to establish a one-to-one correspondence between points on the original closed curves and points on the closed curves of their affine transformed version. Consequently, these closed curves are put on the image of the image, and pixels on these closed curves are derived. Finally, a Fourier transform is conducted on these pixel series. As a result, AIRFDs are derived. Experimental results show that the proposed method can be used for object classification.

Multilayer-Pseudopolar Fractional Fourier Transform Approach for Image Registration

In this paper, a method is proposed to estimate the log-polar coordinates coefficients using Multilayer-peudopolar Fractional Fourier transform (MPFFT). The MPFFT grid is dense at low coefficients and sparse at high coefficients. This character is geometrically similar to the log-polar grid. Large translations, rotations and scaling in images can be estimated. Experimental results show that the proposed method is more efficient for image registration.

Automated flare prediction using the AdaBoost algorithm

We propose a flare prediction method based on the AdaBoost algorithm, which constructs a strong prediction model from a combination of several basic models. Three predictors, extracted from the photospheric magnetograms, are applied as features to predict the occurrence of flares with a certain level over 24 hours following the time when the magnetogram is recorded. To demonstrate the effectiveness of the proposed method, comparisons of experimental results with respect to some existing methods are given. The results show that an improvement is achieved in predicting the occurrences of large flares.

Image Retrieve Based on Primary Clusters

In this paper, a new image retrieval method, which uses the two primary features of image cluster to retrieve image, i.e., the average color value and the number of pixels for each cluster, is proposed. In comparison with popular methods, since this method uses both color feature and structure feature, it is good to solve such a sort of image retrieval questions for two images that they are similar but not identical color. Experimental results show that this method not only enhances the flexibility of system retrieval, but also achieves prominent effects for image retrieval.

Polar radius integral transform for affine invariant feature extraction

Affine transform is to describe the same target at different viewpoints to obtain the relationship between images of approximate model. Affine invariant feature extraction plays an important role in object recognition and image registration. Firstly, the definition of polar radius integral transform (PRIT) is put forward by means of the characterization of affine transform mapping straight line into straight line, where PRIT computes the integral along the polar radius direction and converts images into closed curves which keep the same affine transform with original images. Secondly, in order to extract affine invariant feature, an affine invariant feature extraction algorithm is also given based on PRIT. The proposed algorithm can be used to combine contour-based methods with region-based methods. It has some advantages of fewer amounts of computations and feasibility of feature extraction for objects with several components. Finally, the capability of anti-noise (Gaussian noise, salt and pepper noise) ...

Feature extraction using extended central projection

A novel feature extraction method is proposed in this paper. Dislike contour-based or region-based approaches, an object is first converted to a closed curve by extended central projection (ECP). The derived curve not only keeps the affine transform information, but also is very robust to noise. Then whitening transform is performed to the curve such that the affine transformation is simplified to a rotation only. Finally, Fourier transform are employed to remove the rotation. Several experiments have been conducted to evaluate the performance of the proposed method. Experimental results show that the proposed method has a powerful discrimination ability, and is more robust to noise.

Visibility Estimation Based on Mean Shift

An algorithm is developed to estimate visibility from images captured by video cameras. We propose to smooth the captured image by mean shift and extract the landmark edge by an improved ADM. Then the visibility is estimated in terms of a simplified visibility definition. In comparison with traditional methods, our method not only remove the noise in landmark boundary region, but also make boundary region be more uniform and the extracted edge be more salient. Experimental results illustrate the effectiveness of our approach in visibility estimation.