Takashi Totsuka

Frequency domain volume rendering

The Fourier projection-slice theorem allos projections of volume data to be generated in O(nsquare log n) time for a volumbe of size ncube. The method operates by extracting and inverse Fourier transforming 2D slices from a 3D frequency domain representation of the volume. Unfortunately, these projections do not exhibit the occlusion that is characteristic of conventional volume renderings. We present a new frequency domain volume rendering algorithm that replaces much of the missing depth and shape cues by performing shading calculations in the frequency domain during slice extraction. In particular, we demonstrate frequency domain methods for computing linear or nonlinear depth cueing and directional diffuse reflection. The resulting images can be generated an order of magnitude faster than volume renderings and may be more useful for many applications.

Combining frequency and spatial domain information for fast interactive image noise removal

Scratches on old films must be removed since these are more noticeable on higher definition and digital televisions. Wires that suspend actors or cars must be carefully erased during post production of special effects shots. Both of these are time consuming tasks but can be addressed by the following image restoration process: given the locations of noisy pixels to be replaced and a prototype image, restore those noisy pixels in a natural way. We call it image noise removal and this paper describes its fast iterative algorithm. Most existing algorithms for removing image noise use either frequency domain information (e.g low pass filtering) or spatial domain information (e.g median filtering or stochastic texture generation). The few that do combine the two domains place the limitation that the image be band limited and the band limits be known. Our algorithm works in both spatial and frequency domains without placing the limitations about band limits, making it possible to fully exploit advantages from each domain. While global features and large textures are captured in frequency domain, local continuity and sharpness are maintained in spatial domain. With a judicious choice of operations and domains in which they work, our dual-domain approach can reconstruct many contiguous noisy pixels in areas with large patterns while maintaining continuity of features such as lines. In addition, the image intensity does not have to be uniform. These are significant advantages over existing algorithms. Our algorithm is based on a general framework of projection onto convex sets (POCS). Any image analysis technique that can be described as a closed convex set can be cleanly plugged into the iteration loop of our algorithm. This is another important advantage of our algorithm. CR Categories: I.3.3 [Computer Graphics]: Picture / Image Generation; Display Algorithms; I.3.6 [Computer Graphics]: Methodology and Techniques – Interaction techniques; I.4.4 [Image Processing]: Restoration; I.4.9 [Image Processing]: Applications. Additional

AutoKey: human assisted key extraction

Key extraction is an inverse problem of finding the foreground, the background, and the alpha from an image and some hints. Although the chromakey solves this for a limited case (single background color), this is often too restrictive in practical situations. When the extraction from arbitrary background is necessary, this is currently done by a time consuming manual task. In order to reduce the operator load, attempts have been made to assist operators using either color space or image space information. However, existing approaches have their limitations. Especially, they leave too much work to operators. In this paper, we present a key extraction algorithm which for the first time, addresses the problem quantitatively. We first derive a partial differential equation that relates the gradient of an image to the alpha values. We then describe an efficient algorithm that provides the alpha values as the solution of the equation. Along with our accurate motion estimation technique, it produces correct alpha values almost everywhere, leaving little work to operators. We also show that a careful design of the algorithm and the data representation greatly improves human interaction. At every step of the algorithm, human interaction is possible and it is intuitive. CR Categories: I.3.3 [Computer Graphics]: Picture / Image Generation; I.4.6 [Image Processing]: Segmentation Edge and feature detection; I.4.7 [Image Processing]: Feature Measurement; I.5.2 [Pattern Recognition]: Design Methodology Feature evaluation and selection. Additional

Image blending by feature overwrite

This paper describes a new image compositing algorithm that can properly blend images whose features have a high aspect ratio. Although several techniques have been used in image compositing, none of them works well for images whose features have a high aspect ratio. This compositing algorithm is a simulation on 2D projected space, of the physical phenomena of underlying 3D models. Our algorithm consists of the three steps. First, the images are composited using a multiresolution technique. Second, the features of each image are extracted. Finally, the features are overwritten to the composited image. This method makes it possible to composite hair or fur naturally.

Dual domain interactive image restoration: basic algorithm

This paper describes a new fast, iterative algorithm for interactive image noise removal. Given the locations of noisy pixels and a prototype image, the noisy pixels are to be restored in a natural way. Most existing image noise removal algorithms use either frequency domain information (e.g. low pass filtering) or spatial domain information (e.g median filtering or stochastic texture generation). However, for good noise removal, both spatial and frequency information must be used. The existing algorithms that do combine the two domains (e.g. Gerchberg-Papoulis and related algorithms) place the limitation that the image be band-limited and the band limits be known. Also, some of these may not work well when the noisy pixels are contiguous and numerous. Our algorithm combines the spatial and frequency domain information by using projection onto convex sets (POCS). But unlike previous methods it does not need to know image band limits and does not require the image to be band-limited. Results given here show noise removal from images with texture and prominent lines. The detailed textures as well as the pixels representing prominent lines are created by our algorithm for the noise pixels. The algorithm is fast, the cost being a few iterations (usually under 10), each requiring an FFT, IFFT and copying of a small neighborhood of the noise.

Key extraction by image differentiation

Key extraction is an inverse problem of finding the foreground, the background, and the alpha value from an image and some hints. Although the chromakey solves this for a limited case (single background color), this is often too restrictive in practical situations. When the extraction from an arbitrary background is necessary this is currently done by a time consuming manual task. To solve the problem, an algorithm must address the unique requirements of the key extraction: high resolution, analytically correct alpha values, and easy human interaction. We present an image differentiation method which, for the first time, fulfills these requirements.