International Journal of Image and Graphics

In this article, the new concept of diffusion fields based on partial differential equations is applied to character image processing. Specific diffusion fields are developed according to character image structures and features, depending, on the scope of application. Doing so allows the application of a straightforward one-dimensional numerical scheme to image enhancement, erosion, dilation and thinning. The strength of this approach is the flexibility brought by the diffusion field, which can be defined taking into account specific difficulties of grayscale character images with a minimum of prior information. Thus, the application of the algorithm is shown to be robust to singularity points, the creation of spurious branches, variations in stroke thickness and intensity, multimodality, noise and image background patterns. The resulting enhanced images are noise free with sharp edges and the local typical intensity levels preserved. Thinned characters are connected skeletons located on the ridge of the initial character. Again, the typical intensity of the character and background are preserved.

This article presents an automatic system that takes in grayscale scanned images, which could be mixed text/graphic documents, and performs thresholding and skew detection on the document images. The system consists of two major components; multistage thresholding and skew detection. The proposed skew detection algorithm has no restriction on detectable angle range and does not rely on large blocks of text. It works well on textual document images, graphical images and mixed text and graphic images. The performance of the systems was evaluated using over 60 images that consist of real life documents like envelopes and artificial mixed text/graphic icons. The superior performance of thresholding is clear compared to other techniques from the evaluation. The skew detection algorithm is robust when compared with other methods when very few text lines are present in the document image.

In this paper we propose a new algorithm for the skeletonization of handwritten characters. Unlike traditional skeletonization algorithms that relay only on the configuration of a binary image pixel in deciding whether it is deletable or not, Natural Skeletonization (NS) integrates the gray-level information in this process. The underlying principle here, which stems from the elongated properties of the handwritten characters, is that medial pixels of a handwritten stroke are "naturally" darker than its side pixels. NS consists of three steps: (1) the decomposition step; (2) the thinning step; (3) the reconstruction step. The integration of gray-level information is facilitated by the iterative binarization at equally spaced thresholds, which highlights positional differences between the medial and side pixels of a stroke. The advantage of our approach over existing methods is demonstrated by its ability to prevent the "flooding water" and to prevent the boundary noise from developing extraneous branches. One important aspect of the approach is that it relaxes the skeletonization's dependence on the quality and shape of initial binary pattern. The experimental results indicate that the proposed algorithm substantially improves the skeletonization quality compared to experiments with traditional skeletonization methods.

Binarization of gray scale document images is one of the most important steps in automatic document image processing. In this paper, we present a two-stage document image binarization approach, which includes a top-down region-based binarization at the first stage and a neural network based binarization technique for the problematic blocks at the second stage after a feedback checking. Our two-stage approach is particularly effective for binarizing text images of highlighted or marked text. The region-based binarization method is fast and suitable for processing large document images. However, the block effect and regional edge noise are two unavoidable problems resulting in poor character segmentation and recognition. The neural network based classifier can achieve good performance in two-class classification problem such as the binarization of gray level document images. However, it is computationally costly. In our two-stage binarization approach, the feedback criteria are employed to keep the well binarized blocks from the first stage binarization and to re-binarize the problematic blocks at the second stage using the neural network binarizer to improve the character segmentation quality. Experimental results on a number of document images show that our two-stage binarization approach performs better than the single-stage binarization techniques tested in terms of character segmentation quality and computational cost.

Camera-based imaging is used to digitize printed materials whose shape and size are not suitable for flatbed and Xerox scanning. Because these materials are not physically pressed flat before imaging, the resulting imaged content often appears distorted due to the material's underlying shape. We present a novel approach to correct common image distortions that arise in camera-imaged printed materials. Our approach uses the boundary information of the imaged material to compute a corrective warp to undo the distortion. Our algorithm is unique in that it simultaneously corrects for a variety of distortions including skew, binder curl, and distortions from folds. In addition, 2.5D information about the boundary can be incorporated to compensate for depth distortion.

The aim of "Shape From Texture (SFT)" methods is to obtain three dimensional information out of the monocular view of a scene. In this article, we detail a 2-step SFT computational method. A new local scales extraction technique is used in the first step to compute the local scale of each point of the picture through an interpolation of wavelet values. In the second step, the interpolation of the local scales map helps get a hold of the textured plane's orientation, using the perspective projection model. This method, which is an improvement over Lu's, was tested on synthetic and real textures. Results obtained were consistently better, notably in the case of small slant angles. We also improved the orientation computation by using the vanishing line equation for macrotextures.

An automatic elastic medical image registration approach is proposed, based on image intensity. The algorithm is divided into two steps. In Step 1, global affine registration is first used to establish an initial guess and the resulting images can be assumed to have only small local elastic deformations. The mapped images are then used as inputs in Step 2, during which, the study image is modeled as elastic sheet by being divided into sub-images. Moving the individual sub-image in the reference image, the local displacement vectors are found and the global elastic transformation is achieved by assimilating all of the local transformation into a continuous transformation. The algorithm has been validated by simulated data, noisy data and clinical tomographic data. Both experiments and theoretical analysis have demonstrated that the proposed algorithm has a superior computational performance and can register images automatically with an improved accuracy.

The paper aims to propose a new approach towards human posture reconstruction and animation from monocular video sequences that contain any kind of human postures and movements. This is a way towards low cost motion capture and at the same time it avoids many limitations of those classical methods. A parameterized human skeleton model based on anatomy is adopted where the angular constraints are encoded in the joints. Criterion Function is defined to represent the residuals between feature points in the monocular image and the corresponding points resulted from projecting the human model to the projection plane. By transforming each segment of the human model to achieve the minimum value of the Criterion Function, the proper human posture that resembles the one represented by the monocular image can be generated. Different kinds of adjustments are utilized to adjust the body parts into the proper locations and orientations in 3D space without camera calibration. In order to find the optimal solution effectively in a high-dimensional parameter space by considering all the parameters simultaneously, the method of Genetic Algorithms is proposed. A procedure is developed to recover the whole body posture, and then a human animation system is developed to animate a series of human movements from monocular image sequences, during which information between consecutive frames is considered to improve the accuracy. Our technique makes it feasible to reconstruct any possible human postures, and experimental results from many monocular images and video sequences are encouraging.

We constrain camera motion to a curved line, and capture panoramic images at discrete points along the line. An efficient 3D plenoptic representation is created using the panoramic images. The method does not need to recover geometric models of the scene and the new view is gotten directly from the sampled images of the scene. Moreover, the user can observe significant parallax and lighting changes at a point away from the sampling line. Compared with 4D plenoptic representation, such as lumigraph or light field, the amount of data is much decreased. By combining this 3D plenoptic function with the other plenoptic functions (3D or 4D), a larger scene can be represented. This representation has good storage and computational efficiency, and is very easy to capture. This paper discusses a complete working system, from capturing and construction to rendering of linear mosaics, using synthetic environments and real scenes.

In volume rendering, an important issue in acceleration is to reduce the calculations for occluded voxels. Although this issue has been addressed in the ray casting approach, it is difficult to apply the idea to the projection approach due to uncertain termination conditions. In this paper, we propose a new method to effectively address the exclusion problem in the projection approach, so the rendering process can be accelerated without impairing the rendered image quality. In the rendering process, this new method employs the dynamic screen technique to manage the pixels whose accumulated opacity has not reached 1.0. A ray-cast link at each pixel is set up to record the rendered voxels for the corresponding ray cast from the pixel to intersect. According to the rendered voxels covering the pixels whose accumulated opacity value is below 1.0, visible voxels are selected to render from front to back by the neighboring relationship between the rendered voxels and the voxels to be rendered. Thus, the occluded voxels are dynamically excluded from the loading and rendering processes accurately. Our proposed method can be in general applied to both parallel and perspective projections, using regular and irregular volume datasets. Our experimental results showed that the proposed method can significantly accelerate volume rendering if the data volume has a high percentage of occluded voxels. This method can also perform fairly efficiently for the expensive shading calculations if requested in volume rendering.

This paper presents a novel scheme based on a hybrid NPR/IBMR approach, for providing non-photorealistic virtual environment navigation from images. As a preprocess, we first divide the background and the foreground components in the image and then construct a simple 3D scene model consisting of background model and foreground object models, together with the corresponding non-photorealistic texture maps. Thus, through interactive navigation at runtime, non-photorealistic walk-through images for novel viewpoints can be generated in real time. To reduce visual artifacts in the resulting animation, we propose an effective method for generating multiresolution and view-dependent non-photorealistic texture maps that preserve the stroke size and coherence across different viewing resolutions and angles.