Ju Jia Zou

Skeletonization of ribbon-like shapes based on regularity and singularity analyses

A major problem with traditional skeletonization algorithms is that their results do not always conform to human perceptions since they often contain unwanted artifacts. This paper presents an indirect skeletonization method to reduce these artifacts. The method is based on analyzing regularities and singularities of shapes. A shape is first partitioned into a set of triangles using the constrained Delaunay triangulation technique. Then, regular and singular regions of the shape are identified from the partitioning. Finally, singular regions are stabilized to produce a better result. Experiments show that skeletons obtained from the proposed method closely resemble human perceptions of the underlying shapes.

A deblocking method for BDCT compressed images based on adaptive projections

A deblocking method based on projection onto convex sets (POCS) is proposed to reduce blocking artifacts in compressed images coded by the block discrete cosine transform. The method differs from existing POCS-based methods in three aspects. Firstly, the adjustment of a pixels intensity is determined by local properties of the pixel. Secondly, three locally adaptive constraint sets are introduced to improve deblocking results. Thirdly, the human visual system modeling and relaxed projections are incorporated to make pixels adjust appropriately. The method is tested on typical images with excellent results.

Extracting strokes from static line images based on selective searching

Abstract Extracting the stroke information is important in understanding static line images. An essential issue of stroke extraction by using a thinning process is how to handle intersections where thinning artifacts often occur. In this paper, a novel stroke extraction method based on a “selective searching” technique is proposed. A tree structure is constructed in the vicinity of an intersection so that a correct path of a stroke can be identified by comparing the traveling cost along each candidate path of the tree. The minimum cost path corresponds to the intrinsic trajectory of the stroke at the intersection. A return-cost is calculated for each path so that better performance can be achieved. The new method has been tested extensively with static handwritten numerals extracted from the NIST data base. Experimental results show that the method is effective and reliable.

Efficient skeletonization based on generalized discrete local symmetries

This paper presents an efficient skeletonization method based on generalized discrete local symmetries. A generalized discrete local symmetry is a local symmetry between a contour pixel and a contour segment on the opposite side of the underlying shape. The centerlines of the local symmetries of a shape form the skeleton of the shape. The proposed method is fast compared to two existing skeletonization methods. A speed-up factor of more than 50 can be achieved for high-resolution images. The method is also robust against noise and geometrical transformations, such as rotation and uniform scaling. The method is suitable for skeletonizing high-resolution images where it can be impractical to use other skeletonization techniques, such as thinning and distance transforms, because of the high computational complexity.

Shape skeletonization by identifying discrete local symmetries

Abstract This paper presents a new skeletonization method based on a novel concept — discrete local symmetry. A skeleton obtained from discrete local symmetries approaches the skeleton of the underlying continuous shape if the sampling is dense enough. Discrete local symmetries can be obtained by computing the constrained Delaunay triangulation of the underlying image. Internal triangles of a triangulation are divided into isolated triangles, end triangles, normal triangles and junction triangles. A discrete local symmetry corresponds to an isolated triangle, or an end triangle, or a normal triangle. Several measures are taken to remove skeletonization artifacts and suppress image noise. The proposed method can produce correct centre lines and junctions. It is efficient and robust against noise. The method is suitable for skeletonizing high-resolution images.

Triangle refinement in a constrained Delaunay triangulation skeleton

This paper presents an algorithm with the purpose of improving upon the already successful constrained Delaunay triangulation (CDT) skeletonisation technique. Using such a triangulation to construct a skeleton has proven very effective, that can sometimes, however, produce triangles that do not represent the true nature of the underlying shape. The contour pixels chosen for triangulation are of significant importance, as they determine the triangle edges that define the skeleton. The algorithm described in this paper deals with this problem by inserting new triangulation points in strategic locations in end, normal and junction triangles. Results show that the skeletons produced by this algorithm are accurate, robust against noise and, above all, comply much better with a humans perception of the image than the original triangulation method.

An improved ASIFT algorithm for matching repeated patterns

Image matching is a well researched topic of computer vision. Several new algorithms have been developed in recent times to deal with repetitive pattern matching and affine invariant matching. This paper presents two improvements over the state-of-the-art Affine-Scale Invariant Feature Transform (ASIFT) algorithm. The first improvement enables ASIFT to match repetitive patterns through the use of Graph Transformation Matching. The second increases the accuracy of matching by estimating the transformation between views more precisely. Results show that the proposed method is able to successfully match repetitive patterns such as the checkerboard. An increase in the number of matches can also be seen for matching views under severe affine transformations or projections.

Sub-pixel edge detection for photogrammetry using laplace difference of Gaussian and 4th order ENO interpolation

In modern photogrammetry for structure health monitoring, the detection of retro reflective targets, play a vital role in the accurate measurement of structural objects. However, due to the limitation of the resolution in photogrammetric equipment, a sufficient accuracy may not be achievable. While existing sub-pixel interpolation techniques may be used to overcome this limitation, they may produce unreliable results when presented with weak edge points. This paper presents a new sub-pixel edge detection algorithm namely, the ENO-LDoG method, which incorporates a high-order ENO interpolation scheme and a Laplace of a difference of Gaussian to accurately determine edge points. Experimental results on sample targets show that a higher precision can be achieved when compared to existing methods.

An evaluation of moving shadow detection techniques

Shadows of moving objects may cause serious problems in many computer vision applications, including object tracking and object recognition. In common object detection systems, due to having similar characteristics, shadows can be easily misclassified as either part of moving objects or independent moving objects. To deal with the problem of misclassifying shadows as foreground, various methods have been introduced. This paper addresses the main problematic situations associated with shadows and provides a comprehensive performance comparison on up-todate methods that have been proposed to tackle these problems. The evaluation is carried out using benchmark datasets that have been selected and modified to suit the purpose. This survey suggests the ways of selecting shadow detection methods under different scenarios.

Texture image classification with discriminative neural networks

Texture provides an important cue for many computer vision applications, and texture image classification has been an active research area over the past years. Recently, deep learning techniques using convolutional neural networks (CNN) have emerged as the state-of-the-art: CNN-based features provide a significant performance improvement over previous handcrafted features. In this study, we demonstrate that we can further improve the discriminative power of CNN-based features and achieve more accurate classification of texture images. In particular, we have designed a discriminative neural network-based feature transformation (NFT) method, with which the CNN-based features are transformed to lower dimensionality descriptors based on an ensemble of neural networks optimized for the classification objective. For evaluation, we used three standard benchmark datasets (KTH-TIPS2, FMD, and DTD) for texture image classification. Our experimental results show enhanced classification performance over the state-of-the-art.