Sketch recognition

About: Sketch recognition is a research topic. Over the lifetime, 1611 publications have been published within this topic receiving 40284 citations.

Vision system for licence plate recognition based on neural networks

Abstract: The paper presents an important example of using artificial neural networks in computer vision. Vehicle Number Plate Recognition is a special form of optical character recognition (OCR). Vehicle number plate recognition is a type of technology, mainly software, which enables computer systems to read automatically the registration number of vehicles from digital pictures. We proposed developed methods based on multilayer feed-forward back-propagation algorithm using one hidden layer that is able to recognize numbers and letters in a plate. We also proposed method that is able to find some area with a number plate, which is cut out from the input image and forwarded to neural network application. The performance of the proposed system has been tested on real images.

Sketch recognition via string kernel

Abstract: Sketch recognition is one of the essential step of sketch understanding. Challenge in sketch recognition is the variation and imprecision present in sketch. Free drawing styles of sketching make it difficult to build a robust sketch recognition system. This paper proposes a novel recognition approach that can recognize primitive shapes, as well as combinations of these primitives. The approach is independent of stroke order, number, as well as invariant to size and aspect ratio of sketch. Feature string is used to represent primitives. We defined a similarity measure on these feature strings that counts common substrings in two input strings, which is referred to as the string kernel in the field of kernel methods. Support vector machine(SVM) is then trained with labeled examples to handle the task of classification. The experiment on hand drawn digital circuit diagrams shows that our system can recognize sketching efficiently and robustly.

Survey of Fusion Techniques for Design of Efficient Multimodal Systems

Abstract: Human gesture is a major component for performing intelligent human computer interaction. Human gestures includes different components of visual actions such as motion of hands, facial expression, and torso, motion of eye, speech, to convey meaning and perform actions. In previous works, manual component of gestures is focused. This paper describes multimodal gesture recognition framework, which combines the different groups of features such as hand movement features and facial expression features. In this paper a survey of the work in the fusion of multiple streams of inputs from different modalities to design multimodal system has been described.

Off-line handwritten Thai name recognition for student identification in an automated assessment system

Abstract: In the field of pattern recognition, off-line handwriting recognition is one of the most intensive areas of study. This paper proposes an automatic off-line Thai language student name identification system which was built as a part of a completed off-line automated assessment system. There is limited work undertaken in developing off-line automatic assessment systems using handwriting recognition. To the authors' knowledge, none of the work on the proposed system has been performed on the Thai language. In addition the proposed system recognises each Thai name by using an approach for whole word recognition, which is different from the work found in the literature as most perform character-based recognition. In this proposed system, the Gaussian Grid Feature (GGF) and the Modified Direction Feature (MDF) extraction techniques are investigated on upper and lower contours, loops from full word contour images of each name sample, and artificial neural networks and support vector machine are used as classifiers. The encouraging recognition rates for both feature extraction techniques were achieved when applied on loop, upper and lower contour images (99.27% accuracy rate was achieved using MDF on artificial neural networks and 99.27% using GGF with a support vector machine classifier).

Chinese large-vocabulary name recognition system using character description and syllable spelling recognition

Abstract: The large-vocabulary name recognition technique is one of the challenging tasks in the application of Chinese speech recognition technology. It can be applied on long-list automatic attendant systems and automatic directory assistance systems. A Chinese name has usually two to three characters with each character pronounced as a single syllable. It is a high perplexity task to recognize a word from a long-list of candidates, like more than three hundred thousand unique names in our experiments, given a very short utterance like one to two seconds of speech. Two novel approaches under an interactive framework are proposed in this paper to aid the recognition of a Chinese name: character description recognition (CDR) and syllable spelling recognition (SSR). Together with our robust finite-state recognizer given a graph-structured syllable lexicon for the full names, we achieved a very promising name recognition success rate, 94.5%, in our system-initiative dialogue system.