Youji Iiguni

Image interpolation for progressive transmission by using radial basis function networks

This paper investigates the application of a radial basis function network (RBFN) to a hierarchical image coding for progressive transmission. The RBFN is then used to generate an interpolated image from the subsampled version. An efficient method of computing the network parameters is developed for reduction in computational and memory requirements. The coding method does not suffer from problems of blocking effect and can produce the coarsest image quickly. Quantization error effects introduced at one stage are considered in decoding images at the following stages, thus allowing lossless progressive transmission.

Hierarchical image coding via cerebellar model arithmetic computers

A hierarchical coding system for progressive image transmission that uses the generalization and learning capability of CMAC (cerebellar model arithmetic computer or cerebellar model articulation controller) is described. Each encoder and decoder includes a set of CMACs having different widths of generalization region. A CMAC with a wider generalization region is used to learn a lower frequency component of the original image. The training signals for each CMAC are progressively transmitted to a decoder. Compression is achieved by decreasing the number of training signals for CMAC with a wider generalization region, and by making quantization intervals wider for CMAC with a smaller generalization region. CMACs in the decoder are trained on the training signals to be transmitted. The output is recursively added to the other so that the quality of image reconstruction is gradually improved. The proposed method, unlike the conventional hierarchical coding methods, uses no filtering technique in both decimation and interpolation processes, and has the following advantages: (i) it does not suffer from problems of blocking effect; (ii) the computation includes no multiplication; (iii) the coarsest reconstructed image is quickly produced; (iv) the total number of transmitted data is equal to the number of the original image pixels; (v) all the reconstructed images are equal to the original image in size; (vi) quantization errors introduced at one level can be taken into account at the next level, allowing lossless progressive image transmission.

A Noise Reduction Method Based on Linear Prediction with Variable Step-Size

A noise reduction technique that uses the linear prediction to remove noise components in speech signals has been proposed previously. The noise reduction works well for additive white noise signals, because the coefficients of the linear predictor converge such that the prediction error becomes white. In this method, the linear predictor is updated by a gradient-based algorithm with a fixed step-size. However, the optimal value of the step-size changes with the values of the prediction coefficients. In this paper, we propose a noise reduction system using the linear predictor with a variable step-size. The optimal value of the step-size depends also on the variance of the white noise, however the variance is unknown. We therefore introduce a speech/non-speech detector, and estimate the variance in non-speech segments where the observed signal includes only noise components. The simulation results show that the noise reduction capability of the proposed system is better than that of the conventional one with a fixed step-size.

Speech Enhancement Based on MAP Estimation Using a Variable Speech Distribution

A novel speech enhancement algorithm based on MAP estimation is proposed in this paper. The proposed speech enhancer uses a variable speech spectral distribution adjusted by the sum of power spectral densities. In a speech segment, the variable speech spectral distribution approaches to a Rayleigh density to keep the quality of the enhanced speech. While in a non-speech segment, it approaches to an exponential density so that the proposed speech enhancer reduces the noise strongly. Simulation results show the effectiveness of the proposed method

An adaptive control system design using a memory based learning system

The CMAC is a neural network that imitates the human cerebellum. The CMAC can approximate a wide variety of nonlinear functions by learning, and the learning speed is very fast. However, the CMAC requires a large memory space, because it is based on a table look-up method. Furthermore, a trade-off between the required memory space and the approximation accuracy is required in the choice of the quantization interval. A CMAC that changes quantization intervals adaptively, called a memory-based learning system (MBLS), is designed. The MBLS decreases quantization intervals for areas with larger variation, but it increases those for areas with small variation. This improves the approximation accuracy and reduces the required memory space. Computer simulation results for the modelling and control problems are presented.

Online recognition of handwritten Hiragana characters based upon a complex autoregressive model

An online recognition method for handwritten Hiragana characters is developed based upon a complex AR model. The time delay of the AR model is enlarged so that global attributes of handwritten characters are well incorporated into the model, and a character segmentation technique is developed for performance improvement. A good recognition score has been obtained for two different writers.

Image restoration from a downsampled image by using the DCT

The high-resolution (HR) image restoration from a downsampled low-resolution (LR) image by using the discrete cosine transform (DCT) is proposed. The downsampling process is modeled in matrix form and the DCT is applied to the downsampling matrix to obtain the sparse matrix representation. The restored HR image is then efficiently computed from the LR image by using the sparsity. The distortion caused by the observation noise is theoretically analyzed, and the power of distortion is shown to be closely equal to the variance of the observation noise. Computer simulations show that the proposed method is superior to the cubic spline interpolation in the HR image restoration performance as far as the amount of the additive noise is small.

Sparse image representations with shift-invariant tree-structured dictionaries

In this paper, we introduce shift-invariant sparse image representations using tree-structured dictionaries. Sparse coding is a generative signal model that approximates signals by the linear combinations of atoms in a dictionary. Since a sparsity penalty is introduced during signal approximation and dictionary learning, the dictionary represents the primal structures of the signals. Under the shift-invariance constraint, the dictionary comprises translated structuring elements (SEs). The computational cost and number of atoms in the dictionary increase along with the increasing number of SEs. In this paper, we propose an algorithm for shift-invariant sparse image representation, in which SEs are learnt with a tree-structured approach. By using a tree-structured dictionary, we can reduce the computational cost of the image decomposition to the logarithmic order of the number of SEs. We also present the results of our experiments on SE learning and the use of our algorithm in an image recovery application.

A nonlinear adaptive estimation method based on local approximation

One of the most important problems in signal processing is to estimate the output for a query from the input/output (I/O) data seen so far. This paper presents a nonlinear adaptive estimation method based on the n-nearest neighbor approach. In this method, observed I/O data are stored in a database in the form of a X-dimensional binary digital search trie (k-D trie), and a nonlinear local model to answer each query is derived based on regularization theory. The database contents are efficiently time updated to follow nonstationary data. A storage procedure allowing a simple and efficient update is developed for reduction in processing time and storage requirement. The effectiveness of the proposed method is demonstrated with both simulation data and real speech signals.

Learning of structuring elements for morphological image model with a sparsity prior

This paper presents a learning method of a structuring element for morphological image generative model by using a maximum a posterior (MAP) estimation. Mathematical morphology provides set-theoretic image processing methods. In the morphological processing, an image is approximated as a union of translated and level-shifted structuring elements. The specification of the structuring element is crucial to application of the morphology for image processing tasks. In this paper, we introduce the MAP estimation of the structuring element from an input image for the morphological modeling. Sparse prior density functions of approximation errors and occurrence of the structuring elements are assumed for the learning. The structuring element is optimized to maximize the likelihood that is estimated from the prior density functions. In experiments, we show that the proposed learning method is capable to extract fundamental micro-structures of texture images as the structuring elements.