International Journal of Pattern Recognition and Artificial Intelligence

Thresholding algorithms are being increasingly used in a wide variety of disciplines to objectively discern patterns and objects in micrographs, still pictures or remotely-sensed images. Our experience has shown that three common thresholding algorithms exhibit indeterminacy, in that different operator inputs may lead to very different pattern characterizations. A grayscale image of a soil profile is used to illustrate this phenomemon in the case of the intermeans (IM), minimum error (ME), and Besag's iterated conditional modes (ICM) algorithms. For the illustrative example, the IM algorithm depends only weakly on the starting point of the iterative process — it converges to only two adjacent threshold values. In contrast, the ME algorithm converges to 14 different threshold values plus a segmentation that identifies the entire image as dye, and one that identifies none of it as dye. The ICM algorithm converges to an even wider variety of final segmentations, depending on its starting point. A noniterative modification of the IM and ME algorithms is proposed, providing a consistent method for choosing from among a set of apparently equally-valid segmentations.

The watershed transformation is a primary tool for segmenting a grey-tone image into subsets that are of interest to a visual observer. The resulting image, however, may often appear oversegmented into a large number of tiny regions (basins), most of which are not significant to the problem of domain. In this paper, a method for removing these nonsignificant basins is presented. The notions of relative significance and intrinsic significance are introduced, which lead to the definition of three types of significance for a basin: strong, weak and partial. The merging of a basin with other basins only occurs when the significance of the basin is not strong, and is restricted to suitably selected adjacent basins. The merging is performed by using an iterated process consisting of two phases. The first involves the removal of certain regional minima, and is accomplished by following either a flooding or a digging scheme. The second identifies the basins corresponding to the regional minima remaining in the image and utilizes the watershed transformation. An appropriate selection of the basins to be merged produces a segmented image perceptually close to the original image. The performance of the proposed method is for the case of astronomic images.

In this paper, human faces are detected using the skin color information and the Lines-of-Separability (LS) face model. The various skin color spaces based on widely used color models such as RGB, HSV, YCbCr, YUV and YIQ are compared and an appropriate color model is selected for the purpose of skin color segmentation. The proposed approach of skin color segmentation is based on YCbCr color model and sigma control limits for variations in its color components. The segmentation by the proposed method is found to be more efficient in terms of speed and accuracy. Each of the skin segmented regions is then searched for the facial features using the LS face model to detect the face present in it. The LS face model is a geometric approach in which the spatial relationships among the facial features are determined for the purpose of face detection. Hence, the proposed approach based on the combination of skin color segmentation and LS face model is able to detect single as well as multiple faces present in a given image. The experimental results and comparative analysis demonstrate the effectiveness of this approach.

In this article we propose a polygonal principal curve based nonlinear feature extraction method, which achieves statistical redundancy elimination without loss of information and provides more robust nonlinear pattern identification for high-dimensional data. Recognizing the limitations of linear statistical methods, this article integrates local principal component analysis (PCA) with a polygonal line algorithm to approximate the complicated nonlinear data structure. Experimental results demonstrate that the proposed algorithm can be implemented to reduce the computation complexity for nonlinear feature extraction in multivariate cases.

In a buyer–seller watermarking protocol, a seller should embed a buyer's fingerprinting and seller's watermark into the work before the sale of a work to a buyer. If the work is violated, the fingerprinting and watermark can be extracted to correctly identify who holds the legal copyright of the work and trace the illegal reseller. In this paper, we propose a robust buyer–seller watermarking scheme based on discrete wavelet transformation (DWT) which embeds the buyer's fingerprinting and seller's watermark into an image. Besides, during the extraction phase, the extraction of the fingerprinting and watermark does not require the original image for more feasibility in real application. The proposed scheme has the following advantages. Firstly, the extraction of the fingerprinting and watermark does not require the original image. Secondly, the watermarked image maintains a good visual quality after the fingerprinting and watermark embedding. Thirdly, the proposed scheme is robust against several image processing attacks.

This paper presents an improved framework for voice retrieval of Mandarin broadcast news speech. First, several unsupervised and data-driven approaches for broadcast news transcription were proposed to improve the speech recognition accuracy and efficiency. Then, a multiscale indexing paradigm for broadcast news retrieval was exploited to alleviate the problems caused by the speech recognition errors and the flexible wording structure of the Chinese language. Finally, we used the PDA as the platform and broadcast radio programs collected in Taiwan as the document collection to establish a speech-based multimedia information retrieval prototype system. Very encouraging results were obtained.