Scene based non-uniformity correction in thermal images using Kalman filter

  • Authors:
  • Amir Averbuch;Gabi Liron;Ben Zion Bobrovsky

  • Affiliations:
  • School of Computer Science, Tel Aviv University, Tel Aviv 69978, Israel;Intel Corp. Haifa, Israel;Department of Electrical Engineering - Systems, School of Electrical Engineering, Faculty of Engineering, Tel Aviv Univeristy Tel Aviv 69978, Israel

  • Venue:
  • Image and Vision Computing
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Thermal array detectors, also known as focal-plane arrays (FPA), are a rapidly developing technology and are used in a variety of civil, medical and military applications. The detectors, which are sensitive to radiation in the infrared band, output a high resolution low noise thermal picture. The existence of non-uniformities in the responsitivity of the element array is a severe problem typical to FPA. These non-uniformities result in a fixed pattern ''curtain''-like feature that appear in the image. One of the most common methods to correct non-uniformity is the use of a uniform reference target. This type of non-uniformity correction has a number of disadvantages. The work presented in this paper proposes a new method to calibrate a thermal detector. The proposed correction method is scene based, relying only on the camera's captured video sequence. The algorithm utilizes redundant information achieved from the thermal camera's high sample rate, combined with the camera's motion. The proposed correction algorithm contains two steps: (1) Application of frame registration that compensates for the camera motion. The registration process matches two consecutive frames and produces a residual (difference) frame. Here, we use well-known techniques. (2) Definition of the calibration parameters as a system of linear equations that is solved with the use of a Kalman filter. The Kalman filter tracks the value of each element specific responsitivity value (unknown) through time. Extensive experimental results demonstrate the success of the proposed scheme. The proposed algorithm necessitates modest computational power (ran on a PC with a Pentium III 550MHz processor) due to the sparsity of the involved matrices.