Introduction

PMMWI Camera

This page contains information on the Research Project TIN2013-43880-R, Imágenes milimétricas pasivas: captación, mejora y detección de amenazas funded by the Ministerio de Ciencia e Innovación de España from 2014 to 2016 with a budget of 105.700,00 €.

The research team consists of 6 doctors from the University of Granada and two undergraduates with extensive experience in project theme, a doctor from the Northwestern University (Evanston, Illinois, USA) and a doctor and a undergraduate from companies.

This page will provide information on the project results and publications and the obtained test images and data.

Summary

PMMWI Camera

A passive millimeter-wave (PMMW) imager detects the natural radiation reflected or emitted by the bodies/objects in the scene. This radiation is focused on a detector and is transformed into an electric signal. Millimeter-waves can penetrate through clothing, plastics ands other materials. This is of great use for detecting metallic and non-metallic objects hidden on people. The passive system does not emit radiation and is therefore 100% safe.

There are a number of very important open problems, in the PMMW imaging field, whose solution is based on research and development in processing and information extraction methods for these images. While these problems are encountered in various types of imaging (e.g., visible, multi-spectral, hyper-spectral), PMMW images have their own inherent characteristics making the problem more challenging. PMMW images are typically low-resolution, with little texture and no color information, and of poor quality. Image processing techniques that allow to blindly deconvolve and super-resolve the observed image to assess the content of the real underlying image by either visual inspection or automatic classification/detection, are needed. It is also necessary to blindly deconvolve and super resolve PMMW images that have been taken using the Compressive Sensing paradigm. This paradigm is utilized to reduce the extended acquisition time needed by PMMWI systems. Once the real underlying PMMW image has been obtained through processing, semantic information needs to be extracted from them. More specifically, threats need to be identified.

In this project, blind deconvolution, super resolution and fusion techniques will be developed to obtain, from low quality and low resolution PMMW images and well as from compressed sensing projections, better quality PMMW images. These improved images will be used on threats detection tasks and an interaction loop will be established between processing and threat detection tasks.

The project will have access to PMMW images to be acquired at the facilities of a Spanish manufacturer of millimeter and sub-millimeter-wave (THz) imaging devices, one of the few within Europe. The Spanish company takes part in the project. It will also have access to compressive sensing PMMW observations provided by a team of a top foreign University which also participates in the project. A wide spectrum multidisciplinary team, with university researchers and industry professionals, is then formed.

This project will research on and provide software implementing state of the art processing and detection/classification tasks on passive millimeter images. Results will be published at highly ranked journals and conferences. The software will improve the current security capabilities of passive millimeter systems, which will positively impact the use of such images in security screening.

Read more »

Objectives

The project aims to achieve three general objectives consisting on research and development of algorithms and methdos to:

O1. Process passive millimiter-wave images,
O2. Process images captures using compressive sensing techniques
O3. Threat detection.

To achieve O1 we will work on the research and development of blind deconvolution, super-resolution and image fusion methods, mainly from the Bayesian point of view. To proccess PMMW images captures using compressive sensing techniques and, therefore, have only a set of observations (projections) captured through a sampling matrix, we propose the research and development of superresolution and blind deconvolution methods from these observed projections. For the purpose of detecting threats (O3) in passive millimeter images we propose to create a broad database of these images, the application and adaptation of machine learning methods and the study of the processing/detection interaction to improve the detection system.

Scientific Results

The project is producing results in different aspects:

  • Models and methods of Bayesian inference will improve the quality of PMMW images.
  • Improved images will facilitate visual inspection tasks and the automatic extraction of semantic information.
  • The success rate of semantic extraction methods (threats detection) we propose will improve current methods based on the use of active contours models.
  • The interaction of processing/information retrieval will improve the success rate of the proposed methods.