Oct. 13, 1999:

• N. Saito and R. Coifman, "Local Discriminant Bases," Proceedings of SPIE, vol. 2303, pp. 2-14, Bellingham, Washington, USA, July 1994.
  
  Abstract:
  We describe an extension to the `best-basis' method to construct an orthonormal basis which maximizes a class separability for signal classification problems. This algorithm reduces the dimensionality of these problems by using basis functions which are well localized in time-frequency plane as feature extractors. We tested our method using two synthetic datasets: extracted features (expansion coefficients of input signals in these basis functions), supplied them to the conventional pattern classifiers, then computed the misclassification rates. These examples show the superiority of our method over the direct application of these classifiers on the input signals. As a further application, we also describe a method to extract signal component from data consisting of signal and textured background.
  
  

Oct. 8, 1999:

• J. Lay and L. Guan, "Image Retrieval Based on Energy Histograms of the Low Frequency DCT Coefficients," ICASSP Proceedings, vol. 6, pp. 3009-3012, Phoenix, Arizona, USA, March 1999.
  
  Abstract:
  With the increasing popularity of the use of compressed images, an intuitive approach for lowering computational complexity towards a practically efficient image retrieval system is to propose a scheme that is able to perform retrieval computation directly in the compressed domain. In this paper, we investigate the use of energy histograms of the low frequency DCT coefficients as features for the retrieval of DCT compressed images. We propose a feature set that is able to identify similarities on changes of image-representation due to several lossless DCT transformations. We then use the features to construct an image retrieval system based on the real-time image retrieval model. We observe that the proposed featured are sufficient for performing high level retrieval on medium size image databases. And by introducing transpositional symmetry, the features can be brought to accommodate several lossless DCT transformations such as horizontal and vertical mirroring, rotating, transposing, and transversing.
  
  

Oct. 7, 1999:

• Y. Huang and R. Chang, "Texture Features for DCT-coded Image Retrieval and Classification," ICASSP Proceedings, vol. 6, pp. 3013-3016, Phoenix, Arizona, USA, March 1999.
  
  Abstract:
  The multiresolution wavelet transform has been shown to be an effective technique and achieved very good performance for texture analysis. However, a large number of images are compressed by the methods based on discrete cosine transform (DCT). Hence, the image decompression of inverse DCT is needed to obtain the texture features based on the wavelet transform for the DCT-coded image. This paper proposes the use of the multiresolution reordered features for texture analysis. The proposed features are directly generated by using the DCT coefficients from the DCT-coded image. Comparisons with the subband-energy features extracted from the wavelet transform, conventional DCT using the Brodatz texture database indicate that the proposed method provides the best texture pattern retrieval accuracy and obtains much better correct classification rate. The proposed DCT based features are expected to be very useful and efficient for texture pattern retrieval and classification in large DCT-coded image databases.
  
  

Oct. 5, 1999:

• T. Chang and J. Kuo, "Texture Analysis and Classification with Tree-structured Wavelet Transform," IEEE Transactions on Image Processing, vol. 2, no. 4, pp. 429-441, October 1993.
  
  Abstract:
  One difficulty of texture analysis in the past was the lack of adequate tools to characterize different scales of textures effectively. Recent developments in multiresolution analysis such as the Gabor and wavelet transforms help to overcome this difficulty. In this research, we propose a multiresolution approach based on a modified wavelet transform called the tree-structured wavelet transform or wavelet packets for texture analysis and classification. The development of this new transform is motivated by the observation that a large class of natural textures can be modeled as quasi-periodic signals whose dominant frequencies are located in the middle frequency channels. With the transform, we are able to zoom into any desired frequency channels for further decomposition. In contrast, the conventional pyramid-structured wavelet transform performs further decomposition only in low frequency channels. We develop a progressive texture classification algorithm which is not only computationally attractive but also excellent performance. The performance of our new method is compared with that of several other methods using the DCT, DST, DHT, pyramid-structured wavelet transforms, Gabor filters, and Laws filters.
  
  

Oct. 4, 1999:

• S. Soltane, N. Kerkeni and J. Angue, "Use of Two-dimensional Discrete Cosine Transform for an Adaptive Approach to Image Segmentation," Proceedings of SPIE, vol. 2666, pp. 242-251, Bellingham, Washington, USA, February 1996.
  
  Abstract:
  We are developing a new statistic method based on Two-Dimensional Discrete Cosine Transform 2D DCT. This transform is the focus point of the segmentation process which is discussed here. The basic operations are: (1) The image is transformed from the spatial domain into the frequency domain, (2) The image is treated by blocks. The approach tries to give a detailed description of the image, making easier its following interpretation. It is directed by local indices, using the adapted treatments, selected by criteria of homogeneity and coherence. The method adopted follows a hierarchical step involving three levels. The first level, aims at extracting in a global way, the locations of the edges of the image by eliminating the homogeneous blocks. This treatment is guided by the use of local indices, luminance, energy...etc. The second level, a classification of blocks selected in the first phase according to the similarly criterion of the indices (average, variance, entropy...) and aggregation of these blocks allows to create the areas on which we extract the fine details. Their characterization is made at a very local level. An effective way of characterizing the areas is to focus oneself on the attributes that allow the description and characterization of the texture. However the study of the texture provides on the image very rich additional information. These allow the selection of best treatment (edges extraction operators) according to its efficiency, at the quality of its detection and the fineness of the results. Once all the finer characteristics have been extracted, we merge the results from each area.
  
  

Oct. 1, 1999:

• A. Jain and F. Farrokhnia, "Unsupervised Texture Segmentation Using Gabor Filters," Pattern Recognition, vol. 24, no. 12, pp. 1167-1186, December 1991.
  
  Abstract:
  This paper presents a texture segmentation algorithm inspired by the multi-channel filtering theory for visual information processing in the early stages of human visual system. The channels are characterized by a bank of Gabor filters that nearly uniformly covers the spatial-frequency domain, and a systematic filter selection scheme is proposed, which is based on reconstruction of the input image from the filtered images. Texture features are obtained by subjecting each (selected) filtered image to a nonlinear transformation and computing a measure of 'energy' in a window around each pixel. A square-error clustering algorithm is then used to integrate the feature images and produce a segmentation. A simple procedure to incorporate spatial information in the clustering process is proposed. A relative index is used to estimate the 'true' number of texture categories.
  
  

Sept. 30, 1999:

• J. Strand and T. Taxt, "Local Frequency Features for Texture Classification," Pattern Recognition, vol. 27, no. 10, pp. 1397-1406, October 1994.
  
  Abstract:
  We propose a texture feature extraction method which allows very small training sets, and due to its very local nature is suitable for classification of small textured structures. The method is based on detection of local extremas along a set of direction vectors in the spatial image representation. The local extremas are interpreted as identifying local frequencies, and used to design features corresponding to the basal frequency characteristics amplitude and wavelength. The good texture discrimination ability of these features is demonstrated on two images, and in a quantitative comparison with features from Gabor filtered images and co-occurrence matrix features. Of the three images used in the quantitative comparison, the local frequency features performed best on two of them. We conclude that the new method is a promising feature extraction method for non-stochastic textures, and that it should be further explored.
  
  

Sept. 29, 1999:

• J. Smith and S. Chang, "Transform Features for Texture Classification and Discrimination in Large Image Databases," IEEE International Conference on Image Processing Proceedings, vol. 3, pp. 407-411, November 1994.
  
  Abstract:
  A method for classification and discrimination of textures based on the energies of images subbands is proposed. It is shown that even with this relatively simple feature set, effective texture discrimination can be achieved. It is reported that over 90% correct classification was attained using the feature set in classifying the full Brodatz collection of textures. Moreover, the subband energy-based feature set can be readily applied to a system for indexing images by texture content in image databases, since the feature can be extracted directly from spatial-frequency decomposed image data.
  
  

• M. Dobson, L. Pierce and F. Ulaby, "Knowledge-based Land-cover Classification Using ERS-1/JERS-1 SAR Composites," IEEE Transactions on Geoscience and Remote Sensing, vol. 34, no. 1, pp. 83-99, January 1996.
  
  Abstract:
  Land-cover classification using an ERS-1/JERS-1 composite is discussed in the context of regional and global scale applicability. Each of the orbiting synthetic aperture radar provide somewhat complementary information since data is collected using significantly different frequencies, polarizations and look angles. A conceptual terrain model is developed to show how simple structural features of terrain surfaces and vegetation cover relate data from the sensors. The knowledge-based, conceptual model is incorporated into a classifier which uses hierarchical decision rules to differentiate land-cover classes. Classification levels are based on man-made features, short tall vegetation, woody stems and foliage.
  
  

Sept. 27, 1999:

• F. Castro, J. Amaral and P. Franco, "Invariant Pattern Recognition of 2D Images Using Neural Networks and Frequency-domain Representation," Proceedings of IEEE International Conference on Neural Networks, vol. 3, pp. 1644-1649, June 1997.
  
  Abstract:
  Frequency domain representation of two dimensional gray-level images is used to develop a pattern recognition method that is invariant to rotation, translation and scaling. Frequency domain representation is a natural feature detector that allows the use of only few directions of highest energy as training data for a set of Artificial Neural Networks (ANNs). We developed a new algorithm that uses the spectral information stored in these ANNs to compare a given image with a known pattern, determining the relative translation between them and yielding a measure of their similarity. The representation and method we adopted has the advantage of leaving only the rotation of the object as a free parameter to be determined by the algorithm. We minimize the spectral resolution noise using Spectral Directional Filtering. Our experimental results indicate that the proposed method has excellent discriminating power.
  
  

Sept. 26, 1999:

• D. Xia, H. Li and Y. Qiu, "Combination of SVD and GLCM in Forest Image Recognition," Proceedings of SPIE, vol. 2907, pp. 68-77, Bellingham, Washington, USA, November 1996.
  
  Abstract:
  Feature extraction is the most fundamental and important problem in satellite forest image recognition problem, but how to extract the feature is an important step towards solving recognition problems. The forest image features commonly consist of visual feature, statistical feature, transform field feature, and algebraic feature. The paper uses SPOT remote sensing forest images as samples, SVD (singular value decomposition) and GLCM (gray-level co-occurrence matrix) as the methods of feature extraction, it also compares and analyzes these two methods. The results of comparison show that GLCM is more effective in describing the visual feature and SVD in describing the internal feature. Forest image statistical feature can describe the image macroscopic features, such as the texture feature shape feature, etc. But SVD can describe the image internal features. The SVD spectral features synthesize the features of image at the pixel level. The combination of the two methods can be more effective in the forest image recognition and classification. We use both SVD and GLCM to extract remote sensing forest image features, and by the combination of these two methods we have got a better recognition of the ground surface forest of remote sensing images than before. Our word shows that the forest recognition rate reaches up to 95%.
  
  

• D. Xia, H. Li and Y. Qiu, "SVD Spectral Feature of Image Processing," Proceedings of SPIE, vol. 2904, pp. 549-556, Bellingham, Washington, USA, November 1996.
  
  Abstract:
  Since Golub and Reinsch proposed the singular value decomposition (SVD) algorithm in 1970, SVD first became an effective method to the least square problems. Recently, SVD has been successfully applied in many fields, such as image data compression, feature extraction and so on. This paper discourses on the singular value spectral sequence (SVSS), gives the connotation and application of SVSS in the image processing. We find that SVSS can describe the intrinsic feature of the image: each element of SVSS is the most alike image of the original image in corresponding multidimensional space. Using SVD method, we obtain a set of images corresponding the spectral number. With the increase of spectral number, we can get more details of image, otherwise, the more coarse sketch. It means, the head of SVSS represents the low frequency domain. We use SPOT images as samples and gain some effective SVD spectral features. Finally, we get very good extraction and classification. Compared to the Fourier transform, we think SVD method will be a good method in the field of pattern recognition.
  
  

Sept. 23, 1999:

• H. Yu and W. Wolf, "Scenic Classification Methods for Image and Video Databases," Proceedings of SPIE, vol. 2606, pp. 363-371, Bellingham, Washington, USA, October 1995.
  
  Abstract:
  The problem of scenic image classification is presented in the paper. On considering the specific nature of this problem, we propose a statistically data-based method, the Hidden Markov Model, to solve this problem. We segment an image and use the sequence of segments as the definition of the image; we then train a HMM on a test set of sequences/research/isip/images to establish a classification. We present preliminary results on the use of a 1D HMM for classification of images as either indoor or outdoor.