Indian Journal of Pure & Applied Physics Vol. 47, October 2009, pp. 703-707 Estimation of spectral response of a consumer grade digital still camera and its application for temperature measurement Anagha M Panditrao*& Priti P Rege** *Cummins College of Engineering for Women, **College of Engineering, Pune, 411 052 E-mail: anagha.panditrao@gmail.com Received 26 November 2008; revised 28 July 2009; accepted 27 August 2009 Consumer grade digital still camera has been used for image acquisition of various heat sources and the spectral response of camera has been obtained. Different sources are used to cover the bandwidth ranging from IR to UV. From the results, it has been observed that the response of the digital camera used to capture source images is adequate to cover visible range (400-700 nm). The images of various visible known heat sources are captured using the mentioned camera. From the captured images, various zones of the source are identified using image-processing algorithms. The actual temperature is measured by placing a miniature k type thermocouple in each zone. The measured temperature values are correlated with the respective colours and features computed from the images. The colour temperature correlation is established by applying various analytical techniques. Calibration of the image-based measurement with contact measurement using an established sensor is performed. Keywords: Digital still camera, Spectral response, Non-contact temperature measurement 1 Introduction Temperature measurement is an important requirement in many industrial processes. At present conventional devices and techniques such as contacttype sensor and pyrometers are used to measure the temperature of various sources of heat. With several visible sources of heat such as the furnaces, Bunsen burner, incandescent lamps and oil lamps, the installation of a sensor is difficult. Many a times response time of the sensor is a constraint. Development of charged coupled device (CCD) cameras has opened up the possibility of new techniques employing two-dimensional imaging 1. Digital still cameras (DSCs) have gained significant popularity in recent years 2. With this rapid progression, colour and multispectral properties of images are becoming increasingly crucial to the field of image processing, often extending and replacing previously known gray scale techniques 3. Colour image generation and processing has been growing at an unprecedented rate over the last two decades. Today, scanners, digital cameras, displays and printers are available at relatively inexpensive prices for commercial and consumer s application 4. Most of the light sources emit light as a function of temperature. So, if the image of these sources is acquired by some means and post processing is carried out it is possible to predict the temperature. Digital photography is the easily available fast technique. It is also possible to acquire the image to get total temperature distribution. Applications of the digital cameras that require accurate colour have received a fair amount of research attention over the last few years. Colourimetrically, the DSC has no control over the illumination source. This makes generic characterization of the device much more complex and challenging task. As far as temperature measurement based on imaging is concerned, characterization of camera is essential. There has been only limited work on colorimetrically characterizing DSCs due to its dependency on environment and the proprietary nature of algorithm 4. In the present study, the spectral response of the digital camera is obtained and qualitative visualization for the temperature measurement is done using digital camera. Hence, imaging sources mentioned above to correlate with the temperature is attempted. 2 Experimental Details It is desired that the device bandwidth should be greater than the source bandwidth. It is quiet possible that the spectra that appear the same to the standard human observer may look quiet different to the digital camera. To verify the bandwidth adequacy, it is necessary to plot the camera characteristics 5.
704 INDIAN J PURE & APPL PHYS, VOL 47, OCTOBER 2009 2.1 Design For the experimentation, different sources are used to cover the bandwidth ranging from IR to UV. While designing the setup it is ensured that the effect of stray light is minimized and reflections of the source is avoided. Different sources used to cover the designed bandwidth are: 5W, 12V Lumina make tungsten filament lamp LED: Colour (R, G, B), UV and IR Enclosure length is adjusted to comply with camera minimum distance specifications. The photographs of the set-up used for image capturing are shown in Fig. 1 2.2 Procedure Photographs of monochromatic light source using colour filters (red, green, and blue) of known wavelengths are taken and the source intensity is changed without affecting its white light nature Similar technique is applied for LED array (red, green, blue). The coloured squares used for shade selection are also photographed illuminating with white natural light source avoiding reflection. Number of photographs are taken in manual mode by varying source intensity and changing source and shutter speed. For different sources and filters the procedure is repeated and is spread over four weeks. 2.3 Images captured The images of different filters of known wavelength and LED array are acquired. The images obtained are shown in Figs 2 and 3 respectively. Source: Filtered White Light source 2.4 Image processing It is quiet possible that the perception of the spectra that appear to the standard human observer may be different than that of the digital camera 2. The colour values (red, green, and blue) of the images acquired are separated. The values are compared with standard digital values. Red coloured images are compared with (255, 0, 0), green and blue are compared with (0, 255, 0) and (0, 0, 255) respectively. The processing is carried out using MATLAB software. The images of RGB components obtained for red LED and blue filter based monochromatic light source are shown in Figs 4 and 5 respectively. Similar processing technique is applied to other coloured LEDs and filters. 3 Results and Discussion After processing the images and comparing the colour values with the standard values, results obtained are tabulated. Table 1 presents the values obtained for source LED. The values obtained for different coloured filters are listed in Table 2. The spectral characteristics 6 of the CCD camera used for image capturing are shown in Fig. 6. It is observed that for both the sources the obtained values are fairly matching with the standard values for R, G and B. Fig. 1 Set-up photographs
PANDITRAO & REGE: SPECTRAL RESPONSE OF DIGITAL STILL CAMERA 705 Fig. 2 Filtered white light source images Fig. 3 LED images obtained Fig. 4 RGB components of red LED Fig. 5 RGB components of blue filter
706 INDIAN J PURE & APPL PHYS, VOL 47, OCTOBER 2009 Table 1 RGB values for colour LED Source, Red Green Blue UV Infrared λ nm contribution (650) (530) (470) Red 255 00 60 133 00 Green 82 252 00 00 00 Blue 00 87 255 255 00 Table 2 RGB values for colour filters Source, Red Green Blue λ nm contribution (650) (530) (470) Red 255 00 60 Green 82 255 00 Blue 00 80 255 Fig. 7 Captured images of various sources and flame zones Fig. 6 Spectral characteristics of camera From these characteristics, area under the curve for blue, green and red is calculated, which varies in 1:1.504:1.557 proportion for B, G and R respectively. It is ensured that the camera has adequate bandwidth of operation to cover the visible range (400-700 nm). 3.1 Image data processing There are many applications if the technique is established, especially when images of different sources like wax candle, domestic stoves and Bunsen burners, oil lamp and furnaces are acquired. Using the above-mentioned digital camera, number of images of various sources mentioned above are acquired and the processing is carried out. Fig. 7 depicts the captured images of various sources and the respective flame zones. The images are captured by varying the camera settings and modes, in a dark room. From the captured images, various flame zones are identified by developing image-processing algorithms. Pixel location and its R, G and B colour properties are obtained for individual pixel. The actual temperature is measured in the different flame zones of the sources mentioned above. Considering the flame volume, range and response time of a sensor, a miniature K type thermocouple sensor is used for temperature measurement. The sensor is manipulated in steps of 5 mm in vertical and horizontal directions in the flame and the temperature is measured at around 150 such points. The correlation of measured temperature values and the respective colours and features of the images is obtained. To find out the correlation between colour values (red, green, blue), the independent variables and temperature, the dependent variables, various techniques such as least square method, polynomial fit, linear and non-linear regression are used. The appropriate method should fit the data well with minimum standard deviation. The analysis is done using commercially available standard statistical software. Difference between actual temperature and predicted temperature by each method was worked out. Standard deviation (SD) of the difference for every method was calculated. The method giving minimum standard deviation is selected. The sample colour and temperature readings for a typical image are shown in Table 3. The tabulated readings are from the blue zone of a typical domestic stove image shown in Fig. 7. The graph showing variation of the temperature with the R, G and B colour components is shown in Fig. 8. In images shown in Fig. 2, aperture of the filter is seen though ground glass which is used prior to filter. The concentration is seen at the center. LED shell is observed in Fig. 3. Blue LED is having colourless, clear and transparent shell. So, illumination may contain some other colour pixels. In red LED analysis prominence is seen for red but there is a contribution
PANDITRAO & REGE: SPECTRAL RESPONSE OF DIGITAL STILL CAMERA 707 Table 3 Actual and predicted temperature values Red Green Blue Actual Predicted temperature ( C) temperature ( C) 8 9 29 900 915 12 10 21 950 980 14 12 34 950 985 14 44 98 650 680 15 16 46 1000 1030 16 49 100 650 680 19 18 58 1050 1040 20 22 61 1000 1005 20 34 97 1050 1025 Fig. 8 Temperature variation with R, G and B colour components of green. There is some contribution due to internal source in camera generating date line. On removing date line using setting, this contribution gets eliminated. Colour temperature correlation is obtained. From the graph shown in Fig. 8, it is observed that the increase in the contribution of green colour decreases the temperature significantly. 4 Conclusions From the results obtained it is observed that the digital camera used to capture source images is having adequate bandwidth to cover visible range (400-700 nm) and could be used for faithful measurement of colours through imaging. Various known visible light sources are photographed using a digital camera of known resolution. Calibration of the image-based measurement with contact measurement using an established sensor is performed. It is observed that the image-based measurements agree with the corresponding contact measurements within an error range of about 5%. Digital photography technique for visible heat sources has a good potential for non-contact temperature measurement. References 1 http://www.wartsila-nsd.com 2 Rananath R, Snyder W, Yoo Y & Drew M, IEEE Signal Magazine, 22 (2005) 34. 3 Vrhel M & Trussell H, IEEE Signal Magazine, 22(2005) 14. 4 Vrhel M & Trussell H, IEEE Signal Magazine, 22(2005) 23. 5 Panditrao A & Rege P, Advances in Numerical Methods, Springer US, 11(2009) 249. 6 SONY CCD Image Sensor Company Manual.