Slant correction is a preprocessing technique to improve segmentation and recognition accuracy for handwritten word recognition. All conventional slant correction techniques were performed by the estimation of the average slant angle and the shear transformation. In this paper, a nonuniform slant correction technique for handwritten word recognition is proposed where the slant correction problem is formulated as a global optimal estimation problem of the sequence of local slant angles. The optimal estimation is performed by a dynamic programming based algorithm. From experimental results it was shown that the present technique outperforms conventional uniform slant correction techniques.

This paper presents a survey of elastic matching (EM) techniques employed in handwritten character recognition. EM is often called deformable template, flexible matching, or nonlinear template matching, and defined as the optimization problem of two-dimensional warping (2DW) which specifies the pixel-to-pixel correspondence between two subjected character image patterns. The pattern distance evaluated under optimized 2DW is invariant to a certain range of geometric deformations. Thus, by using the EM distance as a discriminant function, recognition systems robust to the deformations of handwritten characters can be realized. In this paper, EM techniques are classified according to the type of 2DW and the properties of each class are outlined. Several topics around EM, such as the category-dependent deformation tendency of handwritten characters, are also discussed.

A new efficient two-dimensional warping algorithm is presented, in which sub-optimal warping is attained by iterating DP-based local optimization of warp on partially overlapping subplane sequence. From an experimental comparison with a conventional approximation algorithm based on beam search DP, relative superiority of the proposed algorithm is established.

Applications of neural networks are prevailing in speech recognition research. In this paper, first, suitable role of neural network (mainly back-propagation based multi-layer type) in speech recognition, is discussed. Considering that speech is a long, variable length, structured pattern, a direction, in which neural network is used in cooperation with existing structural analysis frameworks, is recommended. Activities are surveyed, including those intended to cooperatively merge neural networks into dynamic programming based structural analysis framework. It is observed that considerable efforts have been paid to suppress the high nonlinearity of network output. As far as surveyed, no experiment in real field has been reported.

Dynamic Programming based elastic pattern matching method called Branched Reference Rubber String Matching was investigated. As in Rubber String Matching, the reference pattern is represented as a sequence of direction specified vectors and the input pattern as two dimensional dot pattern. In order to improve the coping of topological variations in input pattern, the reference patterns allow partial pattern alternatives and misses. Effect on the recognition time is almost negligible. Experimental results show the effectiveness of the proposed algorithm.

Dynamic programming based one dimensional-two dimensional adaptive pattern matching methods were investigated. In these methods, the reference pattern is represented as a sequence of directional vectors and the input pattern as two dimensional dot pattern. The input pattern needs no preskeletization or local feature analysis, and thus stroke order free top-down pattern matching is carried out. As the starting point, Rubber String Matching algorithm using fixed direction vectors was newly investigated. At latter stages, the reference pattern vectors were permitted some freedom in their directions to cope with abrupt aberrations in input pattern line segments, improving the flexibility of matching. Two cases were considered, allowing 45and approximately 20deviations from the vector directions. The 20version gave the best recognition score.

A new dynamic programming (DP) based algorithm for monotonic and continuous two-dimensional warping (2DW) is presented. This algorithm searches for the optimal pixel-to-pixel mapping between a pair of images subject to monotonicity and continuity constraints with by far less time complexity than the algorithm previously reported by the authors. This complexity reduction results from a refinement of the multi-stage decision process representing the 2DW problem. As an implementation technique, a polynomial order approximation algorithm incorporated with beam search is also presented. Theoretical and experimental comparisons show that the present approximation algorithm yields better performance than the previous approximation algorithm.

A new continuous speech recognition algorithm is described, based on word unit reference pattern, dynamic programming and a finite state automaton syntax control. The algorithm is essentially a generalization of two-level DP-matching connected word recognition algorithm. To improve computation and memory efficiency, a forward dynamic programming method is adopted. Also beam search, time-skipping, and table compression techniques are successfully introduced. This algorithm was experimentally implemented into NEC DP-200 speech recognizer, which marked a realtime operation and a 97% sentence accuracy for a 71 word vocabulary and 63 state automaton syntax control task.