Isolated Persian/Arabic word spotting by label embedding

The aim of word spotting,as a particular case of semantic content based image retrieval (cbir),is to find instances of a word query given as an image or text (string) in a document. in this paper,we propose a holistic approach for persian handwritten word spotting. in addition,the proposed method can be used for handwritten word recognition. this is achieved by a combination of label embedding; attribute based classification and common subspace regression. in this subspace,image and string representation of the same word are close together,so it allows for considering recognition and retrieval task as a nearest neighbor problem. unlike existing methods in word spotting,the suggested representation for a word has fixed length and low dimensionality. on the other hand,the feature extraction process is very fast. we used farsa and iranshahr,two common datasets of isolated persian handwritten words,to evaluate the proposed method. the result of experiments for word spotting is promising. the recognition rates for isolated handwritten words in farsa and iranshahr datasets are 100% and 97% respectively.