Fumiaki Takeda

Rotation-invariant neural pattern recognition system with application to coin recognition

In pattern recognition, it is often necessary to deal with problems to classify a transformed pattern. A neural pattern recognition system which is insensitive to rotation of input pattern by various degrees is proposed. The system consists of a fixed invariance network with many slabs and a trainable multilayered network. The system was used in a rotation-invariant coin recognition problem to distinguish between a 500 yen coin and a 500 won coin. The results show that the approach works well for variable rotation pattern recognition.

High speed paper currency recognition by neural networks

In this paper a new technique is proposed to improve the recognition ability and the transaction speed to classify the Japanese and US paper currency. Two types of data sets, time series data and Fourier power spectra, are used in this study. In both cases, they are directly used as inputs to the neural network. Furthermore, we also refer a new evaluation method of recognition ability. Meanwhile, a technique is proposed to reduce the input scale of the neural network without preventing the growth of recognition. This technique uses only a subset of the original data set which is obtained using random masks. The recognition ability of using large data set and a reduced data set are discussed. In addition to that the results of using a reduced data set of the Fourier power spectra and the time series data are compared.

Multiple kinds of paper currency recognition using neural network and application for Euro currency

Up until now, we have developed banking machines for various kinds of paper currency using neural networks. In this paper, we report an enhanced neuro-recognition system to increase the number of recognition patterns using axis-symmetrical mask and two image sensors. One sensors purpose is discrimination for a known image and another one is exclusion for an unknown image. Concretely, we implement the proposed method to an experimental system, which has two sensors, arranged one above and one below the moving banknote. Finally, we apply this proposed method to Euro currency, which will be issued in 2002, using dummy notes. The effectiveness of the proposed method is shown, numerically.

New computational methods for full and subset Zernike moments

The computation of Zernike radial polynomials contributes most of the computation time in computing the Zernike moments due to the involvement of factorial terms. The common approaches used in fast computation of Zernike moments are Kintners, Pratas, coefficient and q-recursive methods. In this paper, we propose faster methods to derive the full set of Zernike moments as well as a subset of Zernike moments. A hybrid algorithm that uses Pratas, simplified Kintners and coefficient methods is used to derive the full set of Zernike moments. In the computation of a subset of Zernike moments, we propose using the combination of Pratas, simplified Kintners, coefficient and q-recursive methods. Fast computation is achieved by using the recurrence relations between the Zernike radial polynomials of successive order without any involvement of factorial terms. In the first and second experiments, we show both the hybrid algorithms take lesser computation time than the existing methods in computing the full set of Zernike moments and a selected subset of Zernike moments which are not in successive sequence. Both hybrid algorithms have been applied in real world application in the classification of rice grains using full set and subset of Zernike moments. The classification performance using optimal subset of Zernike moments is better than using full set of Zernike moments.

Dish extraction method with neural network for food intake measuring system on medical use

We have been engaging development of a food intake measuring system. This system measures amounts of food intake for each dish on the tray. Therefore, this system needs to extract dish image from a tray image. In this paper, we propose a dish extraction method by neural network (NN). We expect that the proposed method can extract dish image efficiently and exactly even if dishes are over-wrapped each other. While, food sometimes causes miss-recognition of the correct position of the extracted dish image, we newly add food rejection algorithm to the dish extraction method. Finally, we show the effectiveness and usability of the improved proposed method with computer simulation using real data.

Banknote recognition by means of optimized masks, neural networks and genetic algorithms

Abstract Previous work by the authors has proposed a banknote recognition system using a neural network (NN) to develop new types of banknote recognition machines. This system is constructed by means of some core techniques. One is a small-scale neural recognition technique using masks. The second is a mask-optimization technique using a genetic algorithm (GA). The last is a neural hardware technique using a digital signal processor (DSP). This paper focuses on and discusses the mask optimization by the GA, which is the second core technique in the neural recognition system. This technique enables the selection of good masks, that can effectively generate the characteristic values of the input image. Further, the effectiveness of this technique is shown not only by the generalization of the NN, but also by a statistical analysis, using the Italian banknotes. Finally, the feasibility and effectiveness of the neural recognition system is shown by using worldwide banknotes.

Thai Banknote Recognition Using Neural Network and Continues Learning by DSP Unit

Nowadays, neural networks (NNs) are widely used in many fields of engineering and the most famous application is pattern recognition. In our previous researches, a banknote recognition system using a NN has been developed for various applications in worldwide banking systems such as banknote readers and sorters. In this paper, a new kind of banknotes, Thai banknotes, are being proposed as the objects of recognition. First, the slab values, which are the digitized characteristics of banknote by the mask set, are extracted from each banknote image. These slab values are the summation of non-masked pixel values of each banknote. Second, slab values are inputted to the NN to execute its learning and recognition process. Third, for commercial usability, the NN algorithm is implemented on the DSP unit in order to execute the continuous learning and recognition. We show the recognition ability of the proposed system and its possibility for self-refreshed function on the DSP unit using Thai banknotes.

A paper currency recognition method by a small size neural network with optimized masks by GA

Compactness, transaction speed, and cost are important design factors when we apply neural networks to commercial products. We propose a structure reduction method for NNs. We adopt slab values which are sums of input pixels as characteristics of the inputs. But there is the possibility of generating the same slab values even when the inputs are different. To avoid this problem, we adopt a mask which covers some parts of the input. This enables us to reflect the difference of input pattern to slab values with masks. Furthermore, we adopt the genetic algorithm (GA) to optimize the masks. We can generate various effective masks automatically. Finally, we show that the proposed method by neuro-recognition with masks can be applied effectively to paper currency recognition machine using the GA.<<ETX>>

Proposal of food intake measuring system in medical use and its discussion of practical capability

In this paper, a food intake measuring system for medical applications is proposed. The system measures the differences of food images between pre-eaten and post-eaten, and accurately calculates the intake of calorie and nutrition. It can be an assistant of dietitians. The whole operation procedures and each component are introduced. The verification experiments of the system performance are also executed.

Characteristics extraction of paper currency using symmetrical masks optimized by GA and neuro-recognition of multi-national paper currency

We have researched a neural network (NN) recognition method and developed a hardware for paper currency. We have proposed a mask concept to extract characteristics of the paper currency. Furthermore, we have adapted a genetic algorithm (GA) to a mask optimization. We propose a unique mask which has a symmetrical masked area against an axis which divides a long side of the currency, equally. We can obtain the same value from both an upright image and an inverse one of the currency through the mask processor using the axis-symmetrical mask. This means these values are invariant to upright and inverse of the currency conveyance. First we show the geometrical meaning of the axis-symmetrical mask and show the procedure of the their optimization by the GA using Japanese, Italian, Spanish, and French currency. Then we show realization of multi-national currency recognition. Finally, we implement this mask on a neuro-banking machine and discuss the effectiveness using a large quantity of the currency.