• ADCMH
• attribute retrieval
• CNN
• computer vision
• deep learning
• face
• image retrieval
• MLP
• neural network
• synthetic vehicle
• vehicle

Attribute-based image retrieval is a type of cross-modal retrieval system, in which data is described by two modalities, an image and an attribute, and the attribute is used as a query to return the image that satisfies it. It can be used in the field of surveillance to simplify the work of human personnel, returning images from a large database that meet certain attributes, without the human personnel having to check each image individually. To build the attribute retrieval system, we use approaches based on deep neural networks, which have the advantage of learning from data to perform a certain task. Specifically, convolutional neural networks (CNN) and multi-layer perceptrons (MLP) are used. In this work, we take into account two different scenarios: attribute retrieval on faces and attribute retrieval on vehicles. For attribute retrieval on faces and attribute retrieval on vehicles, we use an Attribute-based Deep Cross-Modal Hashing (ADCMH) framework, which is composed of two deep neural networks with different architecture. Only for vehicles, in addition to the ADCMH, a simpler approach that uses a single CNN trained as a multi-class classifier on vehicles is tested to perform attribute retrieval.