المستخلص: |
Although not trivial progress has been made in image recognition but it is considered is one of the most critical field of machine learning because of some limitations of what kind of algorithms or methods used for processing. In this work, a powerful algorithm is proposed for image classification. Many robust and efficient models are proposed to increase classification performance and diminish drawbacks inherited from other models. This paper proposes a technique for recognizing Indian numbers using Neural Networks (NN). The main objective for our study was to recognize Indian numbers which have been used in wide applications such as recognizing numbers in car plates or checks in banks. Our models are test and evaluated using not easy and challenging datasets which we created according for some procedures explained in next our sections. Datasets used in this evaluation consists of Indian digit numbers which consist of around 30 thousand samples. In addition to using advanced technique and challenging dataset, a contemporary and efficient methods widely and formerly used are incorporated in our work to expand model efficiency. To the best of our knowledge, comparing with prior work, our models outperform all other former models although all the proposed models are evaluated using challenging dataset created and pre processed for the first time in this work.
|