Young n Woo

Belief Propagation for Stereo Analysis of Night-Vision Sequences

This paper studies different specifications of belief propagation for stereo analysis of seven rectified stereo night-vision sequences (provided by Daimler AG). As shown in [4], Sobel preprocessing of images has obvious impacts on improving disparity calculations. This paper considers other options of preprocessing (Canny and Kovesi-Owens edge operators), and concludes with a recommended setting for belief propagation on those sequences.

A Study on Stereo and Motion Data Accuracy for a Moving Platform

Stereo and motion analysis are potential techniques for providing information for control or assistance systems in various robotics or driver assistance applications. This paper evaluates the performance of several stereo and motion algorithms over a long synthetic sequence (100 stereo pairs). Such an evaluation of low-level computer vision algorithms is necessary, as moving platforms are being used for image analysis in a wide area of applications. In this paper algorithms are evaluated with respect to robustness by modifying the test sequence with various types of realistic noise. The novelty of this paper is comparing top performing algorithms on a long sequence of images, taken from a moving platform.

Effective feature extraction by trace transform for insect footprint recognition

The paper discusses insect footprint recognition. Footprint segments are extracted from scanned footprints, and appropriate features are calculated for those segments (or cluster of segments) in order to discriminate species of insects. The selection or identification of such features is crucial for this classification process. This paper proposes methods for automatic footprint segmentation and feature extraction. First, we use a morphological method in order to extract footprint regions by clustering footprint patterns. Second, an improved SOM algorithm and an ART2 algorithm of automatic threshold selection are applied to extract footprint segments by clustering footprint regions regardless of footprint size or stride. Third, we use a trace transform technique in order to find out appropriate features for the segments extracted by the above methods. The trace transform builds a new type of data structure from the segmented images, by defining functions based on parallel trace lines. This new type of data structure has characteristics invariant to translation, rotation and reflection of images. This data structure is converted into triple features by using diametric and circus functions; the triple features are finally used for discriminating patterns of insect footprints. In this paper, we show that the triple features found by applying the proposed methods are sufficient to distinguish species of insects to a specified degree.

Performance evaluation of binarizations of scanned insect footprints

The paper compares six conventional binarization methods for the special purpose of subsequent analysis of scanned insect footprints. We introduce a new performance criterion for performance evaluation. The six different binarization methods are selected from different methodologically categories, and the proposed performance criterion is related to the specific characteristics of insect footprints having a very small percentage of object areas. The results indicate that a higher-order entropy binarization algorithm, such as proposed by Abutaleb, offers best results for further pattern recognition application steps for the analysis of scanned insect footprints.

An extension of possibilistic fuzzy c-means with regularization

Fuzzy c-means (FCM) and possibilistic c-means (PCM) are the two most well-known clustering algorithms in fuzzy clustering area, and have been applied in many areas with their original or modified forms. However, FCMs noise sensitivity problem and PCMs overlapping cluster problem are also well known. Recently there have been several attempts to combine both of them to mitigate these problems and possibilistic fuzzy c-means (PFCM) showed promising results. In this paper, we propose a modified PFCM using regularization to reduce noise sensitivity in PFCM further. Regularization is a well-known technique to make a solution space smooth and an algorithm noise insensitive. The proposed algorithm, PFCM with regularization (PFCM-R), takes advantage of regularization and further reduce the effect of noise. Experimental results are given and show that PFCM-R is better than existing methods in noisy conditions.

An approach for evaluating robustness of edge operators using real-world driving scenes

Over the past 20 years there have been many papers that compare and evaluate different edge operators. Most of them focus on accuracy and also do comparisons against synthetic data. This paper focuses on real-world driver assistance scenes and does a comparison based on robustness. The three edge operators compared are Sobel, Canny and the under-publicized phase-based Kovesi-Owens operator. The Kovesi-Owens operator has the distinct advantage that it uses one pre-selected set of parameters and can work across almost any type of scene, where as other operators require parameter tuning. The results from our comparison show that the Kovesi-Owens operator is the most robust of the three, and can get decent results, even under weak illumination and varying gradients in the images.

An intelligent system for container image recognition using ART2-Based self-organizing supervised learning algorithm

This paper proposed an automatic recognition system of shipping container identifiers using fuzzy-based noise removal method and ART2-based self-organizing supervised learning algorithm. Generally, identifiers of a shipping container have a feature that the color of characters is black or white. Considering such a feature, in a container image, all areas excepting areas with black or white colors are regarded as noises, and areas of identifiers and noises are discriminated by using a fuzzy-based noise detection method. Noise areas are replaced with a mean pixel value of the whole image and areas of identifiers are extracted by applying the edge detection by Sobel masking operation and the vertical and horizontal block extraction in turn to the noise-removed image. Extracted areas are binarized by using the iteration binarization algorithm, and individual identifiers are extracted by applying 8-directional contour tracking method. This paper proposed an ART2-based self-organizing supervised learning algorithm for the identifier recognition, which creates nodes of the hidden layer by applying ART2 between the input and the hidden layers and improves the performance of learning by applying generalized delta learning and Delta-bar-Delta algorithm between the hidden and the output layers. Experiments using many images of shipping containers showed that the proposed identifier extraction method and the ART2-based self-organizing supervised learning algorithm are more improved compared with the methods previously proposed.

Nucleus Segmentation and Recognition of Uterine Cervical Pap-Smears

The classification of the background and cell areas is very important but difficult problem due to the ambiguity of boundaries. In this paper, the cell region is extracted from an image of uterine cervical cytodiagnosis using the region growing method. Segmented images from background and cell areas are binarized using a threshold value. And the 8-directional tracking algorithm for contour lines is applied to extract the cell area. Each extracted nucleus is transformed to the original RGB space. Then the K-Means clustering algorithm is employed to classify RGB pixels to the R, G, and B channels, respectively. Finally, the Hue information of nucleus is extracted from the HSI models that are transformed using the clustering values in R, G, and B channels. The fuzzy RBF Network is then applied to classify and identify the normal or abnormal nucleus. The result shows that the accuracy of our method is 80% overall and 66% in 5-class problem according to the Bethesda system.

Segmentation of scanned insect footprints using ART2 for threshold selection

In a process of insect footprint recognition, footprint segments need to be extracted from scanned insect footprints in order to find out appropriate features for classification. In this paper, we use a clustering method in a preprocessing stage for extraction of insect footprint segments. In general, sizes and strides of footprints may be different according to type and size of an insect for recognition. Therefore we propose a method for insect footprint segment extraction using an improved ART2 algorithm regardless of size and stride of footprint pattern. In the improved ART2 algorithm, an initial threshold value for clustering is determined automatically using the contour shape of the graph created by accumulating distances between all the spots within a binarized footprint pattern image. In the experimental results, applying the proposed method to two kinds of insect footprint patterns, we illustrate that clustering is accomplished correctly.

Recognition of Shipping Container Identifiers Using ART2-Based Quantization and a Refined RBF Network

Generally, it is difficult to find constant patterns on identifiers in a container image, since the identifiers are not normalized in color, size, and position, etc. and their shapes are damaged by external environmental factors. This paper distinguishes identifier areas from background noises and removes noises by using an ART2-based quantization method and general morphological information on the identifiers such as color, size, ratio of height to width, and a distance from other identifiers. Individual identifier is extracted by applying the 8-directional contour tracking method to each identifier area. This paper proposes a refined ART2-based RBF network and applies it to the recognition of identifiers. Through experiments with 300 container images, the proposed algorithm showed more improved accuracy of recognizing container identifiers than the others proposed previously, in spite of using shorter training time.