Kyoko Kato

High frequency switched capacitor FIR filters using parallel cyclic type circuits

A novel transversal FIR switched capacitor circuit, called the parallel cyclic type, for high-frequency with low power consumption is presented. The circuit is constructed with several sublocks without an operational amplifier in parallel and a summing operational amplifier, in which a single active subblock moves from one subblock to another cyclically. The circuit has a significantly improved high-frequency performance. Consideration is given to the reduction of the number of capacitors and extension of the filter operating frequency range to a higher frequency. Experimental results are given.<<ETX>>

Least-square-based block adaptive prediction approach for lossless image coding

Natural images often consist of many different regions in intensity variation feature. The block adaptive predictors in lossless image coding often show considerably different entropies of block prediction errors between the horizontal scanning and the vertical one, block by block. This paper proposes a block adaptive prediction approach, in which the least-square-based predictors are designed for both horizontal scanning and vertical scanning for each block. Then the sums of the absolute prediction errors for both scanning directions are compared, and the scanning giving lower sum is selected for its block. In the error image, variances of errors for each block are often spread in some wide range. Therefore, in entropy coding stage, each block is classified into several classes by its error variance and the range coders are utilized, class by class, giving lower entropies than that of no classifying.

An algorithm for extracting text strings from comic strips

In recent years it becomes possible to download comic strips and read them on cell phones. Japanese companies such as SONY have started to deliver digitalized comic strips to cell phone users. The market is attractive and promising since cartoons are popular in Japan across generations, while there are plenty of the well-drawn/well-written classical comic strips. Using a cell phone today, it is possible to store a hundred of comic books and display their images one after another in screen like a slid show. For providing the users with more satisfactory service, however, there are still technical problems to be solved. A major problem is the small screen size, which is limited to 2~2.4 inches (320x480 pixels), for the conveneince of portabilities. When a panel of comic strips is displayed on the screen, its text strings are usually too small to recognize. Reading the text strings in voice is a desirable solution to solve the small screen problem, and doing this automatically is a pioneering work. Assuming to embed optical character recognition (OCR) function in cell phones, we attempted voice-reading texts in comics by using existing OCR software. However, sometimes the OCR software could not extract words correctly. When applied directly to a panel of comic strips that contain both drawings and text strings, OCR software miss-recognizes the drawings as words. Most OCRs are developed for extracting words from document images [1], and still have limitation in its performance. To achieve correct and fast voice-reading of the text strings in comic strings, we need firstly to extract the area of text strings as bitmap image, and then to extract words using OCR. Through two years of research, coding and testing, we have constructed an algorithm for extracting the area of text strings automatically out of comic strips. Hereafter, we briefly describe this algorithm. When lettering text strings, different cartoonists usually use words (or kanjis) in different size, font, or languages. They usually enclose the text strings with a speech balloon, which may be drawn in various shapes. Panel borders are sometimes used as part of the speech balloon. Some drawings may be similar in shape to speech balloon. Therefore, it is difficult to extracte text strings by finding its speech balloon. By analyzing a large numbers of comic strips, we found an important common feature. In most comic panels, drawings such as human/animal characters, speech balloons and backgrounds are connected to the surrounding panel borders, but text strings are not. Based on this finding, we have constructed an algorithm, “borders’ connected components deletion”, to extracte text strings in comic strips. We begin with a black and white binary image of a panel. First, we find and label all of the components that are connected to panel borders. Then we replace the pixels of the connected components to white color. As shown in Fig.1-(b), panel borders and most of the drawings can be deleted, while text stings are kept in the intermediate outputs. Still, some drawings may remain; they will be noises to OCR software. For erasing these noises, as the next step, we calculate the size of every remaining connected component. A connected component having a bounding box that is larger in height/width than a specified value (which may be the height/width of one word), is recognized as a noise whose pixels are deleted. Fig. 1-(c) shows the final output images of text strings while having rather few drawings remaining. Such output images were used as inputs to OCR software and this leads to good performance of voice reading of comic strips. We have tested the algorithm by using 160 randomly selected sample panels of famous cartoonists around the world. For about 80% of the samples, processing result is as good as shown in Fig.1, where the whole text strings are standing out, while few noises remaining. The testing shows that the algorithm extracts text strings effectively regardless of the size, font, format, language used in lettering. For a panel within 500x800 pixels, it took 5~12 seconds for a personal computer to extract text strings and display the results. This shows that the algorithm is efficient enough for real time processing. Pre-processing with this algorithm before using OCR software, fast and correct voice reading has been achieved.

Study of photoacoustic three-dimensional imaging by the simultaneous measurement of multiple frequencies method

In this short note, a new technique for three-dimensional (3-D) high-speed imaging of the subsurface crack of an opaque material is proposed. First, the theoretical explanation and principle of simultaneous measurement of multiple frequencies (SMMF) method are given. The 3-D image and the elapsed time of a drill hole beneath the surface of graphite have also been described.

SC FIR interpolation filters using parallel cyclic networks

This paper presents a high frequency switched-capacitor FIR interpolation filter. The circuit has a multi-stage polyphase structure whose component blocks are realized by parallel cyclic networks. Design example of a partial response filter (class IV) is also included.<<ETX>>

Developing software for translation of comic strips

The rapid advances in information technology have helped the internationalization in many fields. For example, the newest comics of artists can be digitalized and delivered to readers over the world at the same time as they are published at home. The potential market of comics is huge because they are popular among generations and customers can download and read comic strips on mobile devices, such as cell phones or iPods. Delivering new comics to customers as soon as possible requires quick and correct translation of the comics. There are two ways of presenting the translated text strings: either to type the text strings under each panel or to erase the original texts within the speech balloons and replace them with the translated ones. The latter would be the preferred option to readers. However, it needs more handwork and time. When Japanese text is translated to Chinese or English changing its presentation from vertical to horizontal would be required. As the translated text may have more words than the original, the speech balloon may not have enough space for the new text. For example, translated English text is often much longer than the width of original balloon. Thus, last words would occur outside the speech balloon. Although many translation softwares are available on the market none of them could deal with the problems in comic translation. In practice, painting software such as Photoshop is often used to erase the original speech balloon and draw a new and larger one. As such, operations are rather difficult and require good experience in using software and drawing comics. Through this research we have developed a software to support comic translation and provide users with specific functions to allow change of text strings and speech balloons. This software has following remarkable functions: Subdivision function: it automatically subdivides a page of comic strips into frames, each with one panel. Speech balloon selection function: when users click any point within a speech balloon, it automatically extracts the outline of the balloon, highlights it, and makes a new layer of the outline in database. Text strings selection function: it enables the user to select a group of text strings by dragging with a mouse. Text strings removal function: When a speech balloon is selected, it automatically erases all the text strings within it. Text string inserting function: When a speech balloon is selected, it places text strings from the keyboard there. Speech balloon extension function: when a speech balloon is selected, it extends all the pixels on the speech balloon’s outline outward. This is done interactively. Each time the user double-clicks mouse, the outline moves outward by one pixel. This is the most unique function of this software. It enables the user to enlarge a speech balloon while keeping its formation unchanged. Our algorithm for this function is based on the previous research for extracting text strings from comics [1]. Since there is usually a white space between texts and speech balloon (or panel frame), we can firstly get Fig.1-(b) which contains almost no texts. By using radiating lines centered in the position of mouse clicking, we can find controlling points on the original speech balloon (Fig.1-(c)). By extending the controlling points outward while keeping the original panel frame unchanged, we can get a new extended speech balloon similar to the original one (Fig.1-(d) to (f)). As shown in Fig. 2, smooth speech balloon is kept smooth, sharp and harsh balloon is still sharp. This is important to make the translated comics having the same mood and quality as the original. For the implementation, we used VC++ for coding and run the software on a notebook computer. We have tested the software for a range and variety of sample panels selected from many famous cartoonists around the world. The testing results show that the software is useful and efficient in translation. Use of it would lead to better and more efficient comic translation.

A plane-based model of four dimensional snowflakes

Four-dimensional (4D) space is not intuitive to human mind. Commonly used examples of 4D space, such as the deep sea and the universe, are difficult for many people to feel or image. Research on the visualization of high dimensional polygons [1] helps people become aware of high dimensional spaces and feel their beauty, complexity and variations.

Evaluation of Modulation Images Using Fixed Equivalent Approximation Method

We previously developed and presented a theory and experimental results for the layered model of vision consisting of the external world, retina and brain. This model was shown to be very effective in evaluating image quality, and provides a means for comparative evaluations of various modulation methods. We have also proposed the equivalence approximation method, by which computation is made easier than the original model. The purpose of this paper is to propose the fixed equivalence approximation method. In this new method the window size is fixed, so there is no need to optimize the window size. This is a practical improvement, as the optimal window size are sometimes another size and may pose difficulty. We performed experiments using five kinds of modulation methods and three kinds of images. The results indicate that the proposed method gives an appropriate measure of objective evaluation that coincides with subjective evaluation, and can, therefore, be an alternative to the original model or equivalence approximation method. Moreover, the present method has an advantage of considerable time saving.