Um método para correção automática de cartão de resposta utilizando processamento digital de imagem
Carregando...
Arquivos
Data
2018-04-27
Autores
Pereira, Pedro Henrique
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Neste trabalho propõe-se um método que envolve diferentes técnicas de processamento digital de imagem para a correção automática de cartão de respostas utilizados em avaliações de múltipla- escolha, vestibulares, concursos ou processos de seleção. O método tem como base um modelo de referência onde são identificados os seguintes elementos do cartão: campos de marcação das opções, número de opção de cada questão e uma imagem de marcação para identificar as extremidades da folha. No trabalho foi aplicada a técnica de Correlação Cruzada Normalizada para identificação da imagem na extremidade da folha. A Transformada de Hough foi utilizada para identificação das áreas onde se realiza a marcação das opções. Para identificar qual a questão que foi marcada utilizou-se a contagem de pixel. Foram realizados testes com 1.154 gabaritos contendo 40 questões, preenchidos no vestibular de uma faculdade particular. O tempo médio para correção de cada cartão foi de 1,39 segundos. A precisão do método foi avaliada utilizando-se 5% dos cartões digitalizados e processados. Basicamente, a avaliação foi realizada conferindo-se visualmente cada cartão e comparando os resultados com os gerados pela correção automática. O índice de acerto com as amostras utilizadas foi de 100%, o que comprova a eficiência do método proposto.
This work proposes a method that involves different techniques of digital image processing for automatic correction of card responses used in multiple-choice evaluations, “vestibular”, contests or selection processes. The method is based on a reference model where the following elements of the card are identified: marking fields of the options, the option number of each question and a marking image to identify the edges of the sheet. In this work, the Normalized Cross-Correlation technique was applied to identify the image at the leaf end. The Hough Transform was used to identify the areas where the options are marked. To identify which question was selected, the pixel count was used. We performed tests with 1.154 templates containing 40 questions that were completed in the entrance examination of a private college. The average time to correct each card was 1.39 seconds. Samples were collected on 5% of the scanned and processed cards by the method to verify their accuracy. The total sampled was visually checked with the results presented by the application of the method and it was verified the percentage of 100% of correct answers in the reading of the markings.
This work proposes a method that involves different techniques of digital image processing for automatic correction of card responses used in multiple-choice evaluations, “vestibular”, contests or selection processes. The method is based on a reference model where the following elements of the card are identified: marking fields of the options, the option number of each question and a marking image to identify the edges of the sheet. In this work, the Normalized Cross-Correlation technique was applied to identify the image at the leaf end. The Hough Transform was used to identify the areas where the options are marked. To identify which question was selected, the pixel count was used. We performed tests with 1.154 templates containing 40 questions that were completed in the entrance examination of a private college. The average time to correct each card was 1.39 seconds. Samples were collected on 5% of the scanned and processed cards by the method to verify their accuracy. The total sampled was visually checked with the results presented by the application of the method and it was verified the percentage of 100% of correct answers in the reading of the markings.
Descrição
Palavras-chave
Reading template, Visão computacional, Correlação cruzada normalizada, Transformada de Hough, Reading template, Computer vision, Standardized cross-correlation, Hough transformation