Ki-min Kang

Automatic red eye correction and its quality metric

The red eye artifacts are troublesome defect of amateur photos. Correction of red eyes during printing without user intervention and making photos more pleasant for an observer are important tasks. The novel efficient technique of automatic correction of red eyes aimed for photo printers is proposed. This algorithm is independent from face orientation and capable to detect paired red eyes as well as single red eyes. The approach is based on application of 3D tables with typicalness levels for red eyes and human skin tones and directional edge detection filters for processing of redness image. Machine learning is applied for feature selection. For classification of red eye regions a cascade of classifiers including Gentle AdaBoost committee from Classification and Regression Trees (CART) is applied. Retouching stage includes desaturation, darkening and blending with initial image. Several versions of approach implementation using trade-off between detection and correction quality, processing time, memory volume are possible. The numeric quality criterion of automatic red eye correction is proposed. This quality metric is constructed by applying Analytic Hierarchy Process (AHP) for consumer opinions about correction outcomes. Proposed numeric metric helped to choose algorithm parameters via optimization procedure. Experimental results demonstrate high accuracy and efficiency of the proposed algorithm in comparison with existing solutions.

Adaptive sharpening of photos

Sharpness is an important attribute that contributes to the overall impression of printed photo quality. Often it is impossible to estimate sharpness prior to printing. Sometimes it is a complex task for a consumer to obtain accurate sharpening results by editing a photo on a computer. The novel method of adaptive sharpening aimed for photo printers is proposed. Our approach includes 3 key techniques: sharpness level estimation, local tone mapping and boosting of local contrast. Non-reference automatic sharpness level estimation is based on analysis of variations of edges histograms, where edges are produced by high-pass filters with various kernel sizes, array of integrals of logarithm of edges histograms characterizes photo sharpness, machine learning is applied to choose optimal parameters for given printing size and resolution. Local tone mapping with ordering is applied to decrease edge transition slope length without noticeable artifacts and with some noise suppression. Unsharp mask via bilateral filter is applied for boosting of local contrast. This stage does not produce strong halo artifact which is typical for the traditional unsharp mask filter. The quality of proposed approach is evaluated by surveying observers opinions. According to obtained replies the proposed method enhances the majority of photos.

Automatic correction of exposure problems in photo printer

In this paper we consider a problem of automatic enhancement of amateur photos in photo printer. The purpose of correction consists of making photos more pleasant for an observer. The photos with various exposure problems and with poorly distinguishable details in shadow areas are analyzed. Our approach is based on contrast stretching and alpha-blending of both brightness of the initial image and estimations of reflectance. For obtaining reflectance estimation a simplified illumination model is used. The luminance is estimated using bilateral filter. Reflectance is estimated using heuristic functions of ratio between brightness of the initial image and estimation of luminance. The correction parameters are chosen adaptively based on histogram analysis. Noise suppression and some sharpening occur during correction. The time and memory optimization issues are considered. Look-up tables and recursive separable bilateral filter are applied to speed up the algorithm. The quality of the algorithm is evaluated by surveying of observers opinions and by comparisons with already existing software and hardware solutions for local shadow correction. The proposed algorithm was implemented into firmware of Samsung dye-sublimation compact photo printer

A restoration method for distorted image scanned from a bound book

When a bound document such as a book is scanned or copied with a flat-bed scanner, there are two kinds of defects in the scanned image; the geometric and photometric distortion. The root cause of the defects is the imperfect contact between the book to be scanned and the scanner glass plate. The long gap between the book center and the glass plate causes the optical path from the surface of the book and the imaging unit(CCD/CIS) to be different from the optimal condition. In this paper, we propose a method for restoring bound document scan images without any additional information or sensor. We correct the bound document images based on the estimation of the boundary feature and background profile. Boundary Feature is obtained after calculating and analyzing the Minimum Boundary Rectangle which encloses the whole foreground contents with minimum size and the extracted feature is used for correcting geometric distortion; de-skew, warping, and page separation. Background profile is estimated from the gradient map and it is utilized to correct photometric distortion; exposure problem. Experimental results show effectiveness of our proposed method.

A channel-dependent color error diffusion method based on distance constraint

It is well known that the homogeneous dot distribution is one of the important features affecting the image quality for B/W error diffusion method. However, when B/W error diffusion method is independently applied to each color channel, homogeneity between channels would not appear in color binary image. Non-homogeneity in color binary image often generates the overlap of the dots between channels. Especially, the overlap of cyan dots and magenta dots is noticeable to human eyes in color highlights. In order to prevent the overlap of cyan dots and magenta dots, this paper modulates the threshold value that makes the distance between dots in cyan channel and magenta channel be equal to the principal distance. Therefore, cyan dots and magenta dots are homogeneously distributed and the overlap of cyan dots and magenta dots can be prevented. The threshold value is increasing or decreasing according to the difference between the principal distance and the minimum distance. In color highlights, the principal distance is adjusted for satisfying the homogeneity between channels and the homogeneity in the respective channel. For the calculation of the minimum distance, this paper describes the 2D-MPOA(two dimensional minor pixel offset array) that is able to calculate the minimum distance efficiently.

Fusion of high dynamic range scene photos

In this paper we propose an effective approach for creating nice-looking photo images of scenes having high dynamic range using a set of photos captured with exposure bracketing. Usually details of dark parts of the scene are preserved in over-exposed shot, and details of brightly illuminated parts are visible in under-exposed photos. A proposed method allows preservation of those details by first constructing gradient field, mapping it with special function and then integrating it to restore lightness values using Poisson equation. Resulting image can be printed or displayed on conventional displays.

Adaptive removal of show-through artifacts by histogram analysis

When scanning a document that is printed on both sides, the image on the reverse can show through with high luminance. We propose an adaptive method of removing show-through artifacts based on histogram analysis. Instead of attempting to measure the physical parameters of the paper and the scanning system, or making multiple scans, we analyze the color distribution to remove unwanted artifacts, using an image of the front of the document alone. First, we accumulate histogram information to find the lightness distribution of pixels in the scanned image. Using this data, we set thresholds on both luminance and chrominance to determine candidate regions of show-through. Finally, we classify these regions into foreground and background of the image on the front of the paper, and show-through from the back. The background and show-through regions become candidates for erasure, and they are adaptively updated as the process proceeds. This approach preserves the chrominance of the image on the front of the papers without introducing artifacts. It does not make the whole image brighter, which is what happens when a fixed threshold is used to remove show-through.

Channel-dependent error diffusion algorithm for dot-off-dot printing

The simplest way of halftoning color images using error diffusion is to apply scalar error diffusion technique to each of color channels independently. When processed independently for C, M, Y and K channels, cyan and magenta dots are often printed at the same pixel location. Such overlaps between cyan and magenta would appear as color noise in highlight area. Thus, it is desirable to minimize dot-on-dot printing of cyan and magenta especially in highlight area. In order to further improve image quality, combined dot distribution of cyan and magenta should be homogeneous. Also, dot distribution of individual color channel should be even. In this paper, tone dependent error diffusion kernels and serpentine direction of processing are employed for homogeneous dot distribution of individual color channel. A decision rule based on updated values of cyan and magenta is applied to achieve dot-off-dot printing. A channel dependent threshold modulation is proposed to improve combined distribution of cyan and magenta. A criterion to measure homogeneity of dot distribution is also proposed.