Non-Uniformity Correction (NUC)

In uncooled infrared modules, a shutter is typically located in front of the detector. This shutter is an essential tool for Non-Uniformity Correction (NUC). But what exactly is NUC?

                                                                           ITG6417                                               ITP612                                                ITZ6417

To understand NUC, we first learn about Infrared Focal Plane Array (IRFPA). The IRFPA is one of the key components of imaging systems. Due to limitations in manufacturing processes and material properties, the response characteristics of each pixel in the focal plane are not consistent, leading to non-uniformity in the infrared image.

For thermal imaging systems based on uncooled vanadium oxide IRFPAs, the non-uniformity of the image primarily originates from three factors: the optical system, the microbolometer array, and the readout circuit.

   1. Optical System: Optical non-uniformities can arise from lens thermal parameter drift, misalignment of the central and optical axes, non-uniform transmittance, and limited effective field of view. These often result in circularly darkened regions near the four corners of the image.

   2. Microbolometer Array: The differences in the optical, electrical, and thermal properties of each pixel in the microbolometer array are the main sources of image non-uniformity. These differences are caused by manufacturing errors, such as variations in the effective area of the detector pixels, pixel resistance, bridge leg width, and bridge surface thickness, which result in inconsistent thermal resistance change characteristics and dark current across the array.

   3. Readout Circuit: In CMOS-based readout circuits, manufacturing limitations cause variations in capacitance and resistance within specific tolerances. Additionally, wiring length and width inconsistencies contribute to differences in integration time, voltage strength, signal delay, and amplification factors, further impacting non-uniformity.

The non-uniformity in infrared images is the cumulative result of these three factors.

Non-Uniformity in IRFPA

The non-uniformity in IRFPA typically appears as Fixed Pattern Noise (FPN), which may take the form of striped or grid-like patterns. These FPN patterns are common in both cooled and uncooled IRFPA modules. Non-uniformity noise can severely degrade the system's imaging quality, so effective Non-Uniformity Correction (NUC) techniques are essential to reduce or eliminate these effects.

When raw data is first acquired from the detector, it can only qualitatively describe the thermal radiation distribution in the scene. However, to convert this data into temperature readings, a quantitative process is required.

Calibration

Calibration can be thought of as a reverse engineering process. Under laboratory conditions, based on blackbody radiation theory, a blackbody is used as a reference source, and thermal imaging data is collected at various temperatures. Based on these radiation data and the known temperature values of the blackbody, a relationship curve between radiation data and temperature is fitted. In practical temperature measurement applications, the target's absolute temperature can then be calculated using this calibration curve and the corresponding grayscale value of the target.

The calibration process establishes the relationship between between a detector pixel's response and temperature. Due to manufacturing  inconsistencies, the response characteristics of each pixel on the detector array are not identical. So how do we determine the mapping relationship for each pixel? It is impractical to calibrate each pixel individually. Additionally, for different lenses, the transmission rate of the central region is significantly higher than that of the edge region. To solve this problem, engineers typically use a shutter on thermal imaging devices to serve as a "calibration" tool.

The Role of the Shutter in NUC

The shutter is typically made of materials with high-emissivity and uniform thermal conductivity. When the shutter blocks the detector, the detector receives thermal radiation from the shutter, ensuring that each pixel outputs the same temperature. Using the shutter, a mapping relationship for the entire focal plane array can be established. However, this mapping relationship is influenced by the surface temperature of the detector. When a detector is first powered on, the temperature fluctuations are significant, resulting in more frequent calibration. After a period of stable operation, the frequency of these calibrations decreases.

Through these two steps, we can establish the relationship between the response of each pixel and the temperature for the detector.

Calibration Algorithms

   1. Calibration-Based Methods: such as Two-Point Correction, Multi-Point Correction, Polynomial Fitting.

Algorithms typically requires obtaining the calibration coefficients beforehand. These coefficients are then read and processed during the calibration process. These methods are highly accurate and relatively simple but lack the ability to adaptively track drift in detector response characteristics. If the drift is significant, recalibration is necessary to update correction coefficients.

   2. Scene-Based Methods: such as Neural Network, Temporal High-Pass Filtering, Constant-Statistics Constraint.

These methods use scene information changes to estimate the gain and offset of the detector pixels in real-time, adapting to the drift in the pixel output. However, most algorithms assume linear pixel responses, making them relatively complex, less accurate, and challenging to implement in hardware.

Two-Point Correction Method

The Two-Point Correction method, commonly referred to as TPC, involves data acquisition at two temperatures (low and high) using the same blackbody. It is one of the earliest and most mature calibration algorithms. Its application relies on two assumptions:

   1. The detector’s response must be linear within the temperature range of interest.

   2. The detector’s response should be stable over time, with minimal random noise influence. This ensures that non-uniformity introduces multiplicative and additive fixed-pattern noise.

For example, when performing Two-Point Correction on a 1×128 linear detector, the non-uniformity correction coefficients consist of 256, comprising 128 multiplicative coefficients and 128 additive coefficients.

Reference:

LYU Lei, ZHANG Xuefeng. Real-time infrared image nonuniformity correction base on FPGA[J]. Laser Infrared, 2011, 41(6): 641-643.
WANG Kun. Research on marching-on in-time scheme and the fast algorithm of time domain integral equation[D]. Chengdu: University of Electronic Science and Technology of China, 2017.