Tienwei Tsai

Dominant Feature Extraction in Block-DCT Domain

Automatically retrieving images through their low-level visual features has become one of the challenging areas of research recently. Among those distinguishing features, the texture features are one of the main themes in content-based image retrieval (CBIR). In this paper, we propose a novel technique to extract dominant features of images in block-DCT domain. The image is first converted to YUV color space and divided into four subblocks. The Y-component in each subblock is then transformed into DCT coefficients, some regions of which characterize different directional texture feature of that subblock. The directional textures in all subblocks are concatenated together as a single feature vector and used for indexing and retrieval of images. The experimental results show that using proper size of block-DCT to emphasize the regional properties of an image while maintaining its global view performs well in CBIR.

Image Retrieval Based on Dominant Texture Features

This paper presents a novel technique that can be used for fast indexing and retrieval of images based on their dominant texture features. Unlike the existing techniques that use computationally intensive texture features for content-based image retrieval, our proposed features are only derived from the DCT coefficients transformed from the Y-component in YUV color space. The dominant texture feature vector is mainly formed with the fundamental properties of global textures. In addition, to employ the fuzzy cognition concepts, our experimental system allows users to easily adjust weights for each individual feature component. Experimental results show that the proposed feature vector is compact with good retrieval accuracy

An Efficient Indexing Method for Content-Based Image Retrieval

In this paper, a 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 features of an image are extracted from its color histograms. In establishing database, quantization technique is applied to quantize the feature vector 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 (or adjacent GCs) as that of the query image is obtained. 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.

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.

Bird information retrieval using fuzzy semantic concepts

This paper describes the bird information retrieval system using the fuzzy semantic concepts. To deal with the uncertainty, fuzzy logic is often used to provide a convenient tool for interfacing linguistic categories with numerical data and for expressing one users preferences in a gradual and qualitative way. Based on this idea, we can develop a fuzzy semantic query model that formulates a query by using a natural language like process. The proposed system has the capability to deal with problems that are beyond the reach of existing key-word-based techniques as it more of a strategy than a precise methodology in handling uncertainty problems. The key feature of this approach are providing a systematic query process in an environment of impression, uncertainty, and partial truth supporting numerous weighting factors for users to achieve a desired target and developing a friendly interface that is easy for users to form a query.

A robust knowledge-based plant searching strategy

This paper presents a knowledge-based plant information retrieval system that is robust to inaccurate and erroneous user queries. First, a knowledge-based genetic algorithm (GA) corrects the erroneous input vectors before these are fed into a back-propagation neural network (BPNN) that performs the actual query. Experimental results show that the strategy achieves a 75% recall rate and 25% precision rate with a cutoff level of 10 under the misjudgment of shapes. Moreover, a fully trained BPNN dynamically adapts to changes in the environment. Due to its robust and simple user interface and portability, the strategy is particularly applicable to educational settings such as outdoor fieldwork in courses on ecology.

An efficient classification approach based on grid code transformation and mask-matching method

In this paper, we present a two-stage classification approach to recognize the characters in the rare books transcribed by ancient calligraphers. The first stage is coarse classification which uses grid code transformation (GCT) method to quantize the most significant discrete cosine transform coefficients into a finite number of grids. On classifying an unknown character, a reduced set of candidate classes can be retrieved from the corresponding grid code. The second stage is fine classification, which uses a statistical mask-matching method to identify the individual target in the set given by the first stage. In the training phase, we generate one positive mask and one negative mask for each distinct class of characters. Therefore, an unknown character can be recognized by finding the prototype character whose masks are best fitted to it. Experiments were conducted for recognizing handwritten characters in Chinese paleography and showed that our approach performs well in this application domain.