Te-Wei Chiang

A Region-Based Image Retrieval Approach Using Block DCT

This paper proposes a region-based image retrieval approach using block discrete cosine transform (BDCT). In our retrieval system, for simplicity, an image is equally divided into four regions and an additional central region with one fourth size of the image. Therefore, an image is represented by five segmented regions, each of which is associated with a feature vector derived from BDCT. Users can select any region as the main theme of the query image. The relevance between a query image and any database image is ranked according to a similar measure computed from the selected regions between two images. For those images without distinctive objects and scenes, users can still select the whole image as the query condition. The experimental results show that our approach is easy to identify main objects and reduce the influence of background in the image, and thus improve the performance of image retrieval.

Fast 2D to 3D conversion based on wavelet analysis

This paper proposes a depth estimation method which converts two-dimensional images of limited depth of field (DOF) into three-dimensional data. The goal is to separate the focused foreground objects from the blurred background objects in an image. Our approach is based on two observations: (1) the focused objects on an image of limited DOF correspond to the objects with high frequency; (2) the high-frequency area of an image appears high energy on its high-frequency wavelet subbands. In our approach, each image is first transformed to the YUV color space; then Y component of the image is further transformed to the wavelet domain. Afterwards, the high-frequency area of an image can be obtained from analyzing the high-frequency wavelet subbands of the image. Finally, binarizatoin and smoothing techniques are applied to find the position of the focused objects on the image. The experimental result demonstrates the effectiveness of our approach.

Content-based image retrieval using grid-based indexing and grey relational analysis

In this paper, an efficient two-stage approach is proposed for content-based image retrieval (CBIR). In establishing the database, the features of an image are extracted from its color histograms and discrete cosine transform (DCT) coefficients. To improve the retrieval performance, the quantization technique is applied to quantize the vector of color histograms such that the feature space is partitioned into a finite number of grids, each of which corresponds to a grid code (GC). At the first stage, a reduced set of candidate images which have the same GC (or adjacent GCs) as that of the query image is obtained. At the second stage, the remaining candidates are examined by using grey relational analysis on the significant DCT coefficients. The experimental results show that the proposed approach leads to a fast retrieval with good accuracy.

A Hierarchical Grid-Based Indexing Method for Content-Based Image Retrieval

In this paper, a hierarchical grid-based indexing method for content-based image retrieval (CBIR) is proposed to improve the retrieval performance. To develop a general retrieval scheme which is less dependent on domain-specific knowledge, the discrete cosine transform (DCT) is employed as a feature extraction method. In establishing database, quantization technique is applied to quantize the DCT coefficients of each database image, such that the feature space is partitioned into a finite number of grids, each of which corresponds to a grid code (GC). On querying an image, a reduced set of candidate images which have the same GC as that of the query image is obtained at varying levels of grid granularity. In the fine matching stage, only the remaining candidates need to be computed for the detailed similarity comparison. The experimental results show that the proposed method leads to a fast retrieval with good accuracy.

Fast 3D image depth map estimation using wavelet analysis

This paper is to propose a method for depth estimation in a 2D image using wavelet analysis. According to our observations, the high frequency components represent focused objects in images of limited depth of field (DOF) and their high frequency wavelet subbands contain high energy. In our approach, each image is first transformed to YUV domain and the Y component is extracted for further analysis. Afterwards, the high frequency bands are derived with wavelet analysis. Through two stages of smoothing and scale manipulation, the depth map data with less error can be used for some 3D display. The experimental result shows that the proposed approach is effective and efficient for depth map estimation.

A FAST TWO-STAGE CONTENT-BASED IMAGE RETRIEVAL APPROACH IN THE DCT DOMAIN

In this paper, a two-stage content-based image retrieval (CBIR) approach is proposed to improve the retrieval performance. To develop a general retrieval scheme which is less dependent on domain-specific knowledge, the discrete cosine transform (DCT) is employed as a feature extraction method. In establishing the database, the DC coefficients of Y, U and V components are quantized such that the feature space is partitioned into a finite number of grids, each of which is mapped to a grid code (GC). When querying an image, at coarse classification stage, the grid-based classification (GBC) and the distance threshold pruning (DTP) serve as a filter to remove those candidates with widely distinct features. At the fine classification stage, only the remaining candidates need to be computed for the detailed similarity comparison. The experimental results show that both high efficacy and high efficiency can be achieved simultaneously using the proposed two-stage approach.

Improving content-based image retrieval with query-candidate relationship

This paper proposes a query-candidate relationship method to improve content-based image retrieval (CBIR). In our approach, each image is first transformed into the YUV color space. Then, the histogram for each component (i.e., luminance Y, blue chrominance U, and red chrominance V) of the image is obtained, which is served as the color feature of the image. To compensate the inherent shortcoming of the color histograms, i.e., without considering the spatial information (such as object location, shape, and texture), the discrete cosine transform (DCT) is applied to extract the spatial features from the Y component of images. In addition, a query-candidate relationship method is further introduced to analyze the mutual similarity between the query image and the candidate images, so as to improve the retrieval. Experimental results show the effectiveness of our approach.

Combination of multiple classifiers for Chinese recognition

Traditional character recognition systems use a single classifier to determine the true class of a given character. However, by using classifiers of different types simultaneously, classification accuracy could be improved. In this paper, we propose a new approach based on majority vote and statistics to support a combined decision among multiple classifiers. First, we find the strengths and weaknesses of all classifiers through the analysis among test characters, templates and classifiers. Then we devise a combination method that can improve classification performance. Experimental results show the effectiveness of our approach.