Ick Hoon Jang

Content-Based Image Retrieval Using Multiresolution Color and Texture Features

In this paper, we propose a content-based image retrieval method based on an efficient combination of multiresolution color and texture features. As its color features, color autocorrelo- grams of the hue and saturation component images in HSV color space are used. As its texture features, BDIP and BVLC moments of the value component image are adopted. The color and texture features are extracted in multiresolution wavelet domain and combined. The dimension of the combined feature vector is determined at a point where the retrieval accuracy becomes saturated. Experimental results show that the proposed method yields higher retrieval accuracy than some conventional methods even though its feature vector dimension is not higher than those of the latter for six test DBs. Especially, it demonstrates more excellent retrieval accuracy for queries and target images of various resolutions. In addition, the proposed method almost always shows performance gain in precision versus recall and in ANMRR over the other methods.

Reduction of blocking artifact in block-coded images using wavelet transform

We propose a simple yet efficient method which reduces the blocking artifact in block-coded images by using a wavelet transform. An image is considered as a set of one-dimensional signals, and so all processing including the wavelet transform are one-dimensionally executed. The artifact reduction operation is applied to only the neighborhood of each block boundary in the wavelet transform at the first and second scales. The key idea behind the method is to remove the blocking component which reveals stepwise discontinuities at block boundaries. Each block boundary is classified into one of shade region, smooth edge region, and step edge region. Threshold values for the classification are selected adaptively according to each coded image. The performance is evaluated for 512/spl times/512 images JPEG coded with 30:1 and 40:1 compression ratios. Experimental results show that the proposed method yields not only a PSNR improvement of about 0.69-1.06 dB, but also a subjective quality nearly free of the blocking artifact and edge blur.

Color Image Enhancement Based on Single-Scale Retinex With a JND-Based Nonlinear Filter

In this paper, we propose a color image enhancement based on the single-scale retinex (SSR) with a just noticeable difference (JND)-based nonlinear filter. In the proposed method, an input RGB color image is transformed into an HSV color image. Under the assumption of white-light illumination, the S and V component images are enhanced. In the enhancement of the V component image, the illumination is first estimated using the JND-based nonlinear filter. The output V component image is then obtained by subtracting some portion of the log signal of the estimated illumination from the log signal of the input V component image. The histogram modeling is next applied to the output V component image. The S component image is enhanced in proportion to the enhanced ratio of the V component image. Finally an output RGB color image is obtained from the enhanced V and S component images along with the original H component image. Experimental results show that the proposed method yields better performance of color enhancement over the conventional histogram equalization and SSR for test color images.

Locally adaptive Wiener filtering in wavelet domain for image restoration

In this paper, a Wiener filtering method in wavelet domain is proposed for restoring an image corrupted by additive white noise. The proposed method utilizes the multiscale characteristics of the wavelet transform and the local statistics of each subband. The size of a filter window for estimating the local statistics in each subband varies with each scale. The local statistics for every pixel in each wavelet subband are estimated by using only the pixels which have a similar statistical property. Experimental results show that the proposed method has better performance over the conventional Lee filter with a window of fixed size.

Iterative blocking artifact reduction using a minimum mean square error filter in wavelet domain

We propose an iterative algorithm for reducing the blocking artifact in block transform-coded images by using a minimum mean square error (MMSE) filter in wavelet domain. An image is considered to be a set of one-dimensional (1-D) horizontal and vertical signals and a 1-D wavelet transform (WT) is utilized in which the mother wavelet is the first-order derivative of a Gaussian-like function. Using an MMSE filter in the wavelet domain the blocking artifact is reduced by removing the component that causes the variance at the block boundary position in the first-scale wavelet domain to be abnormally high compared to those at the other positions and the variances at the positions near the block boundary position in the second-scale wavelet domain to be somewhat high. This filter minimizes the mean square error (MSE) between the ideal blocking component-free signal and the restored signal in the neighborhood of block boundaries in the wavelet domain. The filter also uses local variance in the wavelet domain for pixel adaptive processing. The filtering and the projection onto a convex set of quantization constraint are performed alternately and iteratively. Experimental results show the proposed method yields not only a PSNR improvement of about 0.5-1.07 dB, but also a subjective quality that is nearly free of the blocking artifact and edge blur.

Skew correction of business card images in PDA

We present an efficient algorithm for skew correction of business card images obtained by a PDA camera. -The proposed method is composed of four parts: block adaptive binarization (BAB), stripe generation, skew angle calculation, and image rotation. In the BAB, an input image is binarized block by block so as to lessen the effects of irregular illumination and shadows over the input image. In the stripe generation, character string clusters are generated merging character strings and their inter-spaces, and then only clusters useful for skew angle calculation are output as stripes. In the skew angle calculation, the direction angles of the stripes are calculated using their central moments and then the skew angle of the input image is determined averaging the direction angles. In the image rotation, the input image is rotated by the skew angle. Experimental results show that the proposed method yields correction rates of 100% for business card images in normal conditions and about 90% for those in ill conditions.

Postprocessing in Block-Based Video Coding Based on a Quantization Noise Model *

We present a model of quantization noise in block-coded videos with some assumptions in wavelet domain and propose a postprocessing method to reduce the quantization noise based on the model. A frame of video sequences is considered as a set of one-dimensional (1-D) horizontal and vertical signals. The quantization noise is considered as the sum of the blocking noise and the remainder noise. We model the blocking noise as an impulse or that along with a dispersed impulse at each block boundary in the wavelet domain. The validity of the blocking noise model is investigated. We also model the remainder noise as white Gaussian noise at non-edge pixels in the wavelet domain. Whether the model accommodates well to the remainder noise or not is also examined. The blocking noise is reduced by subtracting a profile, whose strength is adaptively estimated, at each block boundary from the coded signal. The remainder noise then is reduced by a soft-thresholding. We also propose a fast algorithm for the proposed method by approximating coefficients of shape profiles used in blocking noise reduction and inverse wavelet transform (WT) filters used in remainder noise reduction. The performance is evaluated for QCIF video sequences coded by H.263 TMN5 with quantization parameter (QP) in the range of 5--25 and is compared to that of the MPEG-4 verification model (VM) post-filter. Experimental results show that the proposed method yields not only PSNR improvement of maximum 0.5--dB over the VM post-filter but also subjective quality nearly free of the blocking artifact and edge blur.

Texture feature-based language identification using wavelet-domain BDIP, BVLC, and NRMA features

In this paper, we propose a texture feature-based language identification using wavelet-domain BDIP (block difference of inverse probabilities), BVLC (block variance of local correlation coefficients), and NRMA (normalized magnitude) features. The proposed method includes three special operations of NRMA, Donohos soft-thresholding, and variance thresholding. In the proposed method, wavelet subbands are first obtained by wavelet transform from a test image and denoised by Donohos soft-thresholding. BDIP, BVLC, and NRMA operators are next applied to the wavelet subbands. Moments for each subband of BDIP, BVLC, and NRMA are then computed and fused into a feature vector. In classification, a stabilized Bayesian classifier, which adopts variance thresholding, searches the training feature vector most similar to the test feature vector. Experimental results show that the proposed method with the three operations yields excellent language identification even with very low feature dimension.

Block adaptive binarization of business card images in PDA using modified quadratic filter

In this paper, we propose a block adaptive binarization (BAB) using a modified quadratic filter (MQF) to binarize business card images of ill conditions acquired by personal digital assistant (PDA) cameras. In the proposed method, a business card image is first partitioned into blocks of 8×8 and the blocks are then classified into character blocks (CBs) and background blocks (BBs) for locally adaptive processing. Each CB is windowed with 24×24 rectangular window centering around the CB and the windowed blocks are improved by the preprocessing filter MQF, in which the scheme of threshold selection in QF is modified. The 8×8 center block of the improved block is binarized with the threshold. A binary image is obtained tiling each binarized block in its original position. Experimental results show that the quality of binary images obtained by the proposed method is much better than that by the conventional global binarization (GB) using QF. In addition, the proposed method yields about 43% improvement of character recognition rate over the GB using QF.

Texture feature-based language identification using Gabor and MDLC features

In this paper, we propose a texture feature-based language identification using Gabor and MDLC (multi-lag directional local correlation) features. In the proposed method, for a test image, Gabor magnitude images are first obtained by Gabor transform and magnitude operator and MDLC images by MDLC operator. Moments for the Gabor magnitude and MDLC images are then computed and fused into a feature vector. WPCA (whitened principal component analysis) finally searches one of training feature vectors most similar to the test feature vector. Experimental results show that the proposed method yields excellent language identification even with low feature dimension due to a well-matched fusion of Gabor and MDLC features.