Noise suppression in infra red detector arrays.
This invention relates to a method and apparatus for noise suppression in infra red detector arrays. The arrays may have detectors of pyroelectric, resistance bolometers, and cooled CdHgTe materials.
Pyroelectric detector arrays are useful in thermal imaging cameras such as those used by the emergency services. In smoke filled buildings they enable users to negotiate within buildings and locate people and sources of fire. Such cameras focus a series of infra red images onto an array of pyroelectric detector elements, whose collective output is displayed on a video screen for viewing by an operator. In low light levels it becomes important to provide as clear a picture as possible on the screen because the detectors are operating near performance limits. Any noise in the system degrades the observable picture.
In several thermal imagers a defect has been observed; it takes the form of irregular horizontal lines appearing on the screen. This can be distracting or worse. '
The above problem is solved, according to this invention, by measuring outputs from each line of detectors to generate a line noise error signal and subtracting this from the line signal itself.
Statement of invention
According to this invention method of reducing low frequency lines noise from a thermal image processor includes the steps of: imaging thermal energy from a scene onto a detector, reading detector outputs to form a succession of line signals collectively forming a frame, processing each line signal to form a video signal representing a visible image of the thermal scene for display on a screen,
Characterised by the steps of
blanking several detectors at both ends of a line signal to prevent receipt of thermal energy from a scene;
dividing a line signal into a noise processing channel A and a delay channel B;
in the noise processing channel A, time integrate signals from the blanked off detectors and subtract this from the present signal to leave white noise and low frequency noise signals; calculate average line noise level at the left and right blanked off sections the line; calculate the best fit line for the line noise in each line signal taking into account its slope over the complete line period to form a line slope noise signal;
in the delay channel B, delay each line of data by the same time delay as that in the noise processing channel A;
subtract line slope noise signal from the noise processing channel A from the delay channel B line signal to remove the low frequency line noise in each line.
According to this invention a thermal imaging processor for processing a thermal image for display on a screen includes: - means for imaging thermal energy from a scene onto a detector array, means for reading detector outputs to form a succession of line signals, processing means for processing each line signal to form a video signal representing a visible image of the thermal scene for display on a screen,
Characterised by: -
several detectors at both ends of each line in the detector array being blanked to prevent receipt of thermal energy from a scene; means for dividing a line signal into a noise processing channel A and a delay channel B;
means in the noise processing channel A, for time integrating signals from the blanked off detectors and subtract them from a current signal to leave white noise and low frequency noise signals; means for calculating average line noise level at the left and right blanked off detectors; means for calculating the best fit line for the line noise taking into account its slope over the complete line period to form a noise slope line signal;
a delay in the delay channel B for delaying each line of data by the same time . delay as that in the noise processing channel A; means for subtracting line slope data from the noise processing channel A from the delay channel B noise slope line signal to remove the low frequency line noise in each line.
The detector array may be an array of ferroelectric thermal detectors, or resistance bolometers, or a single or group of line photo detectors with optical scanning. The number of detectors blanked off at both ends of each line may be 8, 16, 32 or other convenient digital number to simplify processing.
A sporadic low-frequency noise signal exists at the analogue output of the present generation of UK uncooled ferroelectric arrays. The origin of this noise signal is not known but its effect is observable as a low-frequency first order slope superimposed on to the actual thermal imaging response signals and natural white noise on each video line of the displayed image.
With ferroelectric arrays, thermal image information is extracted by taking the difference between two fields of array information using a thermal reference shutter known as a chopper. This process, termed Image Difference Processing (IDP), removes all common DC offsets and sub frame-rate signal drifts leaving only the differential thermal signal response and white noise. However, a low-frequency noise signal remains and can be readily perceived as fluctuations in the image with a line structure. This can be confirmed by viewing adjacent lines of video information on an oscilloscope trace, revealing that the noise structure has high frequency components but mainly varies over the line period with an approximately first order linear slope.
Although the 'line noise' signal has relatively low amplitude (similar to the white noise) and does not severely degrade the overall image quality, its effect is undesirable and distracting to the viewer. This problem exists on the silicon IC designs for two known thermal detector arrays. The operation of the silicon IC in these array formats has been validated using hybrid detector technology and will eventually enter mass production volumes when the integrated detector technology matures. In order to avoid a costly silicon redesign and revalidation process, it would be preferable to employ a corrective algorithm within the image processing electronics to suppress the effect of the unwanted noise signal.
The method of this invention provides a line noise suppression technique that has been successfully demonstrated using a prototype thermal imaging camera. This technique provides an inexpensive solution to the line noise problem with ferroelectric arrays but could also benefit other imaging systems with similar line noise problems. Thermal cameras employing scanning of an image onto a single detector or a single line of detectors may also benefit from the invention.
Brief description of drawings.
The invention will now be described, by way of example only, with reference to the accompanying drawings of which: -
Figure 1 is a block diagram of ferroelectric array image processing to provide a video output signal;
Figure 2 shows an array of ferroelectric detectors;
Figure 3 is a flow diagram of a line noise suppression algorithm.
Figure 4 is an infra red image using a thermal imaging camera without the noise cancelling algorithm;
Figure 5 is the image of Figure 4 with the algorithm in operation;
Figure 6 is an infra red image using a thermal imaging camera without the noise cancelling algorithm;
Figure 7 is the image of Figure 6 with the algorithm in operation;
Figure 1 shows the processing of a thermal image into a video signal for display on a video screen such a liquid crystal display screen or a cathode ray tube screen. An infra red (IR) scene S is imaged through a lens 1 on to a detector array 2. Typically this will be an x,y matrix of 384 x 288 pyroelectric detectors, each detector forming one pixel in a displayed image. Pyroelectric detectors measure temperature differences. To do this each one is alternately exposed to IR in the scene S and a reference temperature provided by the blades 3 of a rotating shutter, controlled by a chopper motor 4.
Each detector in the array 2 is read out sequentially through a buffer and demultiplex unit 5 to give a string of voltage levels in successive time intervals. Each voltage level represents the amount of IR received by each detector in a frame time. The frame time is the time taken to read the whole array 2. The array output is amplified by a channel gain amplifier 6 so that the dynamic range matches the input range of a following analogue to digital converter (ADC) 7. The digitised signal is then processed in a programmable logic device (PLD) 8 as described below, and converted back to an analogue signal in a DAC 9 for supply to a video monitor 10.
The standard digital processing includes the steps of:
Forming an image difference signal in an image difference processor (IDP) 11 ; this compares the current signal level from a given detector with that of the reference level (a signal emitted from the chopper blades 3) stored in memory 12.
Correcting the gain 13 of each detector to compensate for manufacturing differences in output for a standard measured temperature. Each detector characteristics (obtained by previous calibration) are stored in memory 1 and used to control the gain correction 13.
Video scan converting 15, 6 the signal into a suitable video format for display on a video.
All the above steps are controlled by a microcontroller unit 17 with memory 18.
The algorithm of the present invention operates between the IDP 11 and gain correction 13. This algorithm exploits the fact that the line noise signal varies slowly over the duration of a video line with an approximately first order slope. By taking a measure of the noise offset at either end of the video line it is possible to compute the theoretical first order slope of the line noise over the intervening data points. The computed slope can then be subtracted from each data pixel on the active portion of the line, restoring the thermal image information to a normalised average.
The algorithm requires a small section of the left and right sides of the active array 20 to be masked to prevent differential thermal image information breaking through and thus offsetting the algorithm, see Figure 2 where 8 detectors 21 , 22 at both ends of each line are masked. Other numbers could be chosen, e.g. 16 (twice) detectors, but these decrease the number of active detectors and reduce horizontal field of view.
In laboratory experiments masking was achieved using a simple cardboard optical mask positioned in front of the focal plane giving an unchopped thermal signal in those areas with no differential response. The optical mask solution is not ideal however since a small amount of shadowing is observed at the edges of the mask, requiring a larger area to be obscured for correct operation of the algorithm (and thus a greater reduction in the active field of view). The preferred masking solution may be achieved electrically by altering one of the metalisation layers of the silicon readout IC during fabrication to effectively short-circuit the left and right sections of the pixel array to the reset voltage state.
The algorithm steps are shown in Figure 3 and operate on a line signal one line at a time.
Step 1. Divide a LineN raw pixel data signal into a noise processing channel A and a delay channel B. This line signal is the IDP 11 output of each detector in a line with each line read sequentially. The time taken to read all detectors in one line (including blanked detectors) is a line time, and time for all lines to be read is a frame time.
In channel A:
Step 2. Time integrate 25 each detector output. This provides an average amplitude value for each detector.
Step 3. Subtract 26 time integrated detector value from current value to give a signal having white noise and low frequency (LF) noise only.
Step 4. Calculate 27 the average line noise level at the left and right blanked- off sections of the line by summing together the noise signals from the 'blind' eight detectors and then divide by eight to remove the white noise. Output a left and a right LF noise offset signal. Note that this process can be performed with a simple binary divide (i.e. right shift 3) rather than employing complicated division arithmetic logic.
Step 5. Calculate 28 the best fit line for the line noise taking into account its slope over the complete line period to form a line slope noise signal. Output a Line -ι LF noise slope signal.
In channel B:
Step 6. Delay 29 each line of data as they pass through the noise slope calculation algorithm and feedback through the main processing pipeline for correction. Output a
raw pixel data signal.
Step 7. Combine 30 channels A and B and subtract the slope line noise component from each data point on the active portion of the line. Output a LineN-1 pixel data with LF noise suppression.
Step 8. Repeat for every line in the current field, a complete frame.
Steps 1-8 are repeated for successive frames.
The effect of the algorithm is shown in Figures 4-7. Figure 4 is a thermal image of a person processed without use of the algorithm. Several horizontal lines can be seen particularly across the shirt of the person. This is a snapshot, the position of the lines change with time because they represent noise in the system. Figure 5 is a thermal image with the algorithm in operation; note that the horizontal lines are now missing.
Figure 6 is a thermal image of electrical leads hanging on a wall. Note several horizontal lines. Figure 7 is the same image as in Figure 6 but with the algorithm in operation; note the lack of horizontal lines.
The algorithm may be implemented in hardware, e.g. in a programmable logic device or frozen in an A.S.I.C. design, or instructional software steps incorporated in a recording and storage medium such as a memory chip or disk.