Detecting Buildings in Historical Photographs Using Bag-of-Keypoints

Abstract

Abstract-The strategies for the preservation of historical documents can include their digitization, which is an effective way to make them publicly available while preventing degradation of the original sources. The Arquivo Público Mineiro (APM), the Archives of the State of Minas Gerais, has a collection of historical photographs from Brazil, and some of them have been digitized. The availability of digital copies of historical photographs makes it possible to apply ContentBased Image Retrieval (CBIR) techniques to alleviate the huge manual effort that is put nowadays into their description and indexing. On the other side, such images are usually more challenging than modern photographs, because of the poor quality of the originals and several degradation effects.In this work, it is proposed a technique based on a bag-of-keypoints representation to identify images containing buildings in the APM photographic collection. The bag-ofkeypoints is an efficient image representation technique, which has been proved robust to occlusion and variations due to pose, scale, illumination and several transformations. Experiments were performed on the images from the APM collection, to classify them between building and non-building, using bagof-keypoints representations of those images. Results show that, despite of the poor quality of the images, the bag-ofkeypoints representation is able to provide good detection rates, indicating the suitability of the proposed method for the task.

Non-locally Enhanced Feature Fusion Network for Aircraft Recognition in Remote Sensing Images

“…Csurka et al [24] assign patch descriptors to a set of predetermined clusters, construct a bag of keypoints, which counts the number of patches assigned to each cluster, and then train multi-class classifiers using the bags of keypoints as feature vectors. Batista et al [25] propose a technique based on a bag-of-keypoints representation to identify images containing buildings in the APM photographic collection. The self-correlation matrix of the image is a simple and straightforward method to model the correlation between long-distance pixels.…”

Detecção de cenas em segmentos semanticamente complexos

“…No entanto, não resolvem o problema dos segmentos semanticamente complexos (Lopes e Goularte, 2013), uma vez que os descritores locais não possuem o poder de generalização necessário para diferenciar categorias com diferentes semânticas (Valle e Cord, 2009). Situação semelhante (discriminação x generalização) vem sendo enfrentada na área de Recuperação de Informação Baseada em Conteúdo (Content-Based Information Retrieval -CBIR) com o uso de técnicas Bag of Visual Words, obtendo bons resultados em categorização semântica latente de imagens (Sivic e Zisserman, 2006;Batista et al, 2009;Lazebnik et al, 2006;Ionescu et al, 2013). Essa técnica consiste em extrair descritores visuais das imagens, para a formação de um vocabulário, que é utilizado para determinar imagens semanticamente relacionadas, de modo latente.…”

Caracterização e recuperação de imagens usando dicionários visuais semanticamente enriquecidos

“…Desde então outros trabalhos foram propostos para os mais diversos domínios. O trabalho de(Batista et al, 2009), por exemplo, explora a técnica BoVW com o objetivo de detectar construções em fotografias históricas;(Lopes et al, 2009) para a detecção de nudez em cenas de vídeo;…”