Seong-O Shim

A proposal for evolutionary fuzzy systems using feature weighting: dealing with overlapping in imbalanced datasets

In a general scenario of classification, one of the main drawbacks for the achievement of accurate models is the presence of high overlapping among the concepts to be learnt. This drawback becomes more severe when we are addressing problems with an imbalanced class distribution. In such cases, the minority class usually represents the most important target of the classification. The failure to correctly identify the minority class instances is often related to those boundary areas in which they are outnumbered by the majority class examples. Throughout the learning stage of the most common rule learning methodologies, the process is often biased to obtain rules that cover the largest areas of the problem. The reason for this behavior is that these types of algorithms aim to maximize the confidence, measured as a ratio of positive and negative covered examples. Rules that identify small areas, in which minority class examples are poorly represented and overlap with majority class examples, will be discarded in favor of more general rules whose consequent will be unequivocally associated with the majority class. Among all types of rule systems, linguistic Fuzzy Rule Based Systems have shown good behavior in the context of classification with imbalanced datasets. Accordingly, we propose a feature weighting approach which aims at analyzing the significance of the problems variables by weighting the membership degree within the inference process. This is done by applying a different degree of significance to the variables that represent the dataset, enabling to smooth the problem boundaries. These parameters are learnt by means of an optimization process in the framework of evolutionary fuzzy systems. Experimental results using a large number of benchmark problems with different degrees of imbalance and overlapping, show the goodness of our proposal.

Accurate shape from focus based on focus adjustment in optical microscopy.

Optical microscopy allows a magnified view of the sample while decreasing the depth of focus. Although the acquired images from limited depth of field have both blurred and focused regions, they can provide depth information. The technique to estimate the depth and 3D shape of an object from the images of the same sample obtained at different focus settings is called shape from focus (SFF). In SFF, the measure of focus–sharpness–is the crucial part for final 3D shape estimation. The conventional methods compute sharpness by applying focus measure operator on each 2D image frame of the image sequence. However, such methods do not reflect the accurate focus levels in an image because the focus levels for curved objects require information from neighboring pixels in the adjacent frames too. To address this issue, we propose a new method based on focus adjustment which takes the values of the neighboring pixels from the adjacent image frames that have approximately the same initial depth as of the center pixel and then it re‐adjusts the center value accordingly. Experiments were conducted on synthetic and microscopic objects, and the results show that the proposed technique generates better shape and takes less computation time in comparison with previous SFF methods based on focused image surface (FIS) and dynamic programming. Microsc. Res. Tech., 2009.

Edge color histogram for image retrieval

The classical color histogram for image indexing does not take into account the shape information of an image. A color histogram method with edge information is studied. Color distributions are found for the pixels of three types of edges (two directional edges and one non-directional edge) and three distance measures are computed on the basis of the color distribution of each edge type. The proposed similarity measure obtained by combining these three distance measures could reduce the false match rate in comparison with using only single (non-directional) edge types. Simulation results show an improvement in indexing quality as compared to that of the traditional color histogram and edge histogram.

Image indexing by modified color co-occurrence matrix

Image indexing based on modified color co-occurrence matrix (MCCM) is proposed in this paper. First, CCM is simplified to represent the number of color (hue) pairs between adjacent pixels in the image. And then, CCM is split into diagonal and non-diagonal elements that constitute two elements of MCCM. Indexing the image by MCCM could exploit shape information in abstract level. Proposed MCCM is accumulative feature. Experimental results show the superiority of the proposed MCCM based indexing in comparison to the indexing based on other accumulative features, color histogram and auto-correlogram with very competitive computational cost.

A novel iterative shape from focus algorithm based on combinatorial optimization

Shape from focus (SFF) is a technique to estimate the depth and 3D shape of an object from a sequence of images obtained at different focus settings. In this paper, the SFF is presented as a combinatorial optimization problem. The proposed algorithm tries to find the combination of pixel frames which produces maximum focus measure computed over pixels lying on those frames. To reduce the high computational complexity, a local search method is proposed. After the estimate of the initial depth map solution of an object, the neighborhood is defined, and an intermediate image volume is generated from the neighborhood. The updated depth map solution is found from the intermediate image volume. This update process of the depth map solution continues until the amount of improvement is negligible. The results of the proposed SFF algorithm have shown significant improvements in both the accuracy of the depth map estimation and the computational complexity, with respect to the existing SFF methods.

Depth Map Estimation using a Robust Focus Measure

Accurate estimation of depth map leads to precise three-dimensional shape recovery. In this paper, we present a new focus measure for calculation of depth map. This new focus measure is based on an optical transfer function implemented in the frequency domain and it has shown robustness in the presence of noise as compared to the earlier focus measures. The results of the proposed focus measure have shown considerable improvement in the presence of noise with respect to other focus measures.

Shape from focus using principal component analysis in discrete wavelet transform

We introduce a new approach for 3-D shape recovery based on discrete wavelet transform (DWT) and principal component analysis (PCA). A small 3-D neighborhood is considered to incorporate the effect of pixels from previous as well as next frames. The intensity values of the pixels in the neighborhood are then arranged into a vector. DWT is applied on each vector to decompose it into approximation and wavelet coefficients. PCA is then applied on modified energies of wavelet components. The first feature in the eigenspace, as it contains maximum variation, is employed to compute the depth. The performance of the proposed approach is tested and is compared with existing methods by using synthetic and real image sequences. The evaluation is gauged on the basis of unimodality and monotonicity of the focus curve. Resolution, accuracy, root mean square error (RMSE), and correlation metrics have been applied to evaluate the performance. Experimental results and comparative analysis demonstrate the effectiveness of the proposed method.

EEG based evaluation of stereoscopic 3D displays for viewer discomfort

BackgroundConsumer preference is rapidly changing from 2D to 3D movies due to the sensational effects of 3D scenes, like those in Avatar and The Hobbit. Two 3D viewing technologies are available: active shutter glasses and passive polarized glasses. However, there are consistent reports of discomfort while viewing in 3D mode where the discomfort may refer to dizziness, headaches, nausea or simply not being able to see in 3D continuously.MethodsIn this paper, we propose a theory that 3D technology which projects the two images (required for 3D perception) alternatively, cannot provide true 3D visual experience while the 3D technology projecting the two images simultaneously is closest to the human visual system for depth perception. Then we validate our theory by conducting experiments with 40 subjects and analyzing the EEG results of viewing 3D movie clips with passive polarized glasses while the images are projected simultaneously compared to 2D viewing. In addition, subjective feedback of the subjects was also collected and analyzed.ResultsA higher theta and alpha band absolute power is observed across various areas including the occipital lobe for 3D viewing. We also found that the complexity of the signal, e.g. variations in EEG samples over time, increases in 3D as compared to 2D. Various results conclude that working memory, as well as, attention is increased in 3D viewing because of the processing of more data in 3D as compared to 2D. From subjective feedback analysis, 75% of subjects felt comfortable with 3D passive polarized while 25% preferred 3D active shutter technology.ConclusionsWe conclude that 3D passive polarized technology provides more comfortable visualization than 3D active shutter technology. Overall, 3D viewing is more attractive than 2D due to stereopsis which may cause of high attention and involvement of working memory manipulations.

Depth from focus based on combinatorial optimization

Depth from focus (DFF) is a technique to estimate the depth and 3D shape of an object from a sequence of images obtained at different focus settings. The DFF is presented as a combinatorial optimization problem. After the estimate of the initial depth map solution of an object, the algorithm updates the depth map iteratively from the specially defined neighborhood. The results of the proposed DFF algorithm have shown significant improvements in both the accuracy of the depth map estimation and the computational complexity, with respect to the existing DFF methods.

Image de-noising using noise ratio estimation, K-means clustering and non-local means-based estimator

One of the key issues in removing random-valued impulse noise from digital images using switching filters is the impulse noise detection. Impulse noise is a random, spiked variation in the brightness of the image. In this paper, a new impulse noise detection algorithm is presented that is based on Noise ratio Estimation and a combination of K-means clustering and Non-Local Means based filter (NEK-NLM). Luo-statistic is employed as a non-local means based estimator. The novelty of this work lies in the introduction of a pre-processing step of noise ratio estimation before noise detection, this estimation allows us to select suitable parameters for the noise detection algorithm. In noise filtering stage, nonlocal-means estimator is applied for restoring noisy pixels to their actual values. Using real world datasets, this paper shows that the impulse noise can be removed effectively. Extensive comparison of simulation results with the already published results show that the proposed method outperforms most of the existing impulse noise removal techniques both in terms of noise detection and image restoration.