|Year : 2017 | Volume
| Issue : 3 | Page : 116-120
Development of cluster count algorithm for radiation measurement using complementary metal oxide semiconductor camera
Aatef Shaikh, Mukesh K Sharma, MS Kulkarni, Jis Romal, Ashutosh Gupta, Probal Chaudhury
Radiation Safety Systems Division, Bhabha Atomic Research Centre, Mumbai, Maharashtra, India
|Date of Submission||29-May-2017|
|Date of Decision||19-Jun-2017|
|Date of Acceptance||12-Jul-2017|
|Date of Web Publication||16-Feb-2018|
Radiation Safety Systems Division, Bhabha Atomic Research Centre, Trombay, Mumbai - 400 085, Maharashtra
Source of Support: None, Conflict of Interest: None
Complementary metal oxide semiconductor (CMOS) camera was used as radiation detector. The methodology to use CMOS active pixel sensor for radiation monitoring purpose is discussed. A new cluster count algorithm to measure dose rates from the images captured using the commonly available CMOS cameras in cell phones and tablets for gamma radiation measurement was developed and implemented on personal computer (PC). Images were taken in standard gamma fields and analyzed on PC to generate the calibration coefficients. The algorithm was validated using 60Co source for dose rates up to 10 Gy/h by images captured on Samsung Galaxy GT-6800 tablet. The algorithm was also tested using other radiation sources such as 137Cs, 241Am, and X-rays for different dose rates.
Keywords: Cluster count algorithm, complementary metal oxide semiconductor pixel, radiation dose rate
|How to cite this article:|
Shaikh A, Sharma MK, Kulkarni M S, Romal J, Gupta A, Chaudhury P. Development of cluster count algorithm for radiation measurement using complementary metal oxide semiconductor camera. Radiat Prot Environ 2017;40:116-20
|How to cite this URL:|
Shaikh A, Sharma MK, Kulkarni M S, Romal J, Gupta A, Chaudhury P. Development of cluster count algorithm for radiation measurement using complementary metal oxide semiconductor camera. Radiat Prot Environ [serial online] 2017 [cited 2020 Feb 21];40:116-20. Available from: http://www.rpe.org.in/text.asp?2017/40/3/116/225580
| Introduction|| |
Complementary metal oxide semiconductor (CMOS) active pixel sensors find its use in many applications including cell phone camera, webcams, pocket cameras, and digital single-lens reflex (DSLRs). A basic sensor contains an array of pixels each having a photosensitive diode, an amplifier, and a control circuit for controlling the operation of the sensor. Like visible light photons, gamma-rays are capable of producing electron–hole pairs in the CMOS detector material; therefore, the CMOS camera can be used as radiation detector. When a gamma-ray interacts with photodetector in a CMOS pixel, energy of the gamma-ray is partially transferred to the pixel creating electron–hole pairs in the detector material. The amount of energy transferred and hence the charge generated depends on the mass-energy attenuation coefficient of silicon for the respective energy of gamma photon and the thickness of the material. In radiation environment, pixel detector produces signal corresponding to both visible light and gamma photon. Hence, during the experiments, the photodetector was covered using an opaque tape to block the visible light while gamma-rays still reach the photodetector.
The use of smartphone camera as a gamma radiation detector has been reported by many researchers., It can serve as a useful device to the first responders in case of nuclear and radiological emergencies. In case of such emergency situations, the best radiation detector is the one, which is available at the moment. Presently specialized radiation monitoring instruments are used to fulfill these purposes. However, it is not feasible to put these specialized instruments everywhere. The first responders need information of affected area, prevailing dose rate in the affected area and type of radionuclide. Smart cell phones with camera are commonly available with the general public and can provide the estimate of dose rate and other information in such emergency situations. In addition to the built-in camera, the biggest advantage of smartphone/tablet is its processing power. An android application can be developed for using the camera for radiation detection and measurement.
Cogliati et al. has reported an algorithm using CMOS camera in a cell phone for gamma detection. The CMOS pixels generate considerable noise signal due to leakage current and ambient temperature conditions and must be subtracted from the input image. The Cogliati et al. algorithm generates signal image by applying high-delta method to remove noise in the input image. The dose rate was measured based on the number of signal image pixels having value more than a particular threshold and multiplying that number with a suitable multiplier. However, this algorithm saturates at dose rates of about 1 Gy/h. This paper proposes a new cluster count algorithm to overcome the saturation issue faced in this algorithm and to increase the range of dose rate measurements. The objective of this work is to develop a new methodology to use CMOS cameras available in cell phones and tablets for radiation measurement that can be used in the event of radiation emergencies.
| Methodology|| |
The pixel size in modern high-resolution CMOS imager is typically of the order of 1–4 μm. A gamma photon entering pixel array will interact with an electron and imparts its energy to free an electron. This recoil electron will travel in the detector material creating electron–hole pairs along its path, in the process losing its energy. This process illuminates a cluster of neighboring pixels as shown in [Figure 1]. The count of number of such clusters in an image can be correlated to the gamma dose rate in which the image was captured.
|Figure 1: Pixel clusters as a result of gamma photons with Cs-137 source (a) dose rate = 400 mGy/h (b) dose rate = 2.6 Gy/h|
Click here to view
Color image sensors are basically monochrome image sensors having color filters to filter-specific light component (R, G, and B). Gamma photons pass preferentially unaffected by these color filters because of their high energy as compared to visible light photons. Hence, the color information generated by the red, green, and blue components uniformly represents the energy deposited in the pixel. Hence, instead of directly operating on color image, it is converted to a grayscale image by adding red, green, and blue components of the image as per equation 1. Typically, when there is no gamma field present, a dark image is obtained. However, in the presence of a gamma source, bright pixel clusters are observed. In this context, a cluster is considered as a bunch of illuminated pixels connected together. To define connectivity between pixels, a set of intensity values is used. Two neighboring pixels are said to be connected if the intensity values of both belong to V, where V is a set of intensity values. Set V contains all the values above a pixel intensity threshold, which separates background pixels from illuminated pixel clusters. A binary image is generated from the grayscale image by assigning a value “1” for the pixels which have values belonging to set V and a value of “0” for remaining pixels as per equation 2
Where C is the color image, G is the grayscale image, B is the binary image, X and Y are pixel location, and V is a set of all intensity values greater than the threshold.
The flood fill algorithm is used to identify the clusters. This algorithm is called recursively with a seed value. Every time a cluster is identified, a variable called “cluster count” is incremented, and a new seed value is determined. The algorithm labels all the pixels in the identified cluster with the current value of cluster count. The whole process is repeated by traversing the binary image and invoking the flood fill algorithm in a loop for assigning a cluster number to each identified cluster. When the complete image is traversed, the last cluster number is the total cluster counts in the binary image. [Figure 2] and [Figure 3] illustrate input and output image segments of this algorithm.
The first step in identifying the clusters in the captured images is to decide the pixel intensity threshold. For this purpose, variation of cluster counts is observed and plotted with respect to varying pixel intensity threshold as shown in [Figure 4]. It shows the variations of cluster count when the threshold is changed from 0 to maximum pixel intensity value, i.e., 755 in steps of 10. When threshold is 0, all the pixels have a value greater than threshold, and the complete image is counted as 1 big cluster. When the threshold is increased, gradually the background pixels start disappearing and number of clusters increases. After a particular point, the number of clusters starts decreasing with threshold as some clusters not having enough brightness disappear. At one point, the graph becomes almost flat. A point on the flat portion of the graph is chosen as threshold because, at this point, most of the clusters formed due to noise are discarded, and the cluster count becomes relatively independent of variations in threshold.
| Results and Discussion|| |
The high-delta algorithm as described by Cogliati et al. was implemented, and the signal pixel counts were computed for various dose rates using forty images per computation. Similarly, counts were computed using the proposed cluster count algorithm for the identical dose rates. The cluster count algorithm operates on a single image; hence, it takes around five seconds per measurement whereas high-delta algorithm takes around 3 min per measurement. [Figure 5] shows the graph of signal pixel count and cluster count versus dose rate. For high-delta algorithm, it is observed that with increasing dose rate, the pixel count increases initially, and above ~3.5 Gy/h, the algorithm starts to saturate. It is also observed that for very high-dose rates, the signal pixel counts decrease. Whereas the cluster count algorithm is linear up to dose rate of 10 Gy/h, it is also found that at lower dose rates, the cluster counts obtained by cluster count algorithm are comparable to the background images. Hence, this algorithm may be more suitable for dose rates above 5 mGy/h.
|Figure 5: Variation of output of high-delta and cluster count with dose rate|
Click here to view
When trying to fit a single equation relating cluster counts to dose rates, large error is observed at lower dose rates. Hence, data are divided in two ranges, and calibration equation is generated separately for both ranges. The equations are given in the following format:
Where x is “cluster count.”
[Table 1] gives the two cluster count ranges and applicable calibration coefficients obtained for these ranges. Type of the fit and number of decimal places in coefficients has been chosen to keep error below 10% in the computed dose rate. The algorithm and the calibration function are validated using images taken in known gamma field. The results for some of the dose rates are given in [Table 2]. It was observed that the percentage standard deviation associated with the measurement decreases as dose rate increases [Figure 6] which is consistent with the fact that emission of gamma-rays follows Poisson statistics. As dose rate increases, cluster count also increases which results in decrease in percentage standard deviation. It does not decrease below a certain limit due to noise floor associated with the signal and accompanying electronics.
|Table 1: Coefficients of calibration function relating dose rate to counts for cluster count algorithm|
Click here to view
|Figure 6: Variation in standard deviation of measured dose rate for various dose rates|
Click here to view
Tests were conducted in the calibration facility using Cs-137, Co-60, Am-241, and X-rays to obtain the energy response of the algorithm. It is plotted in [Figure 7] for a dose rate of 50 mGy/h. It is observed that for same gamma field, dose rates measured at lower energies are several times higher than those measured for higher energies. Major reason for this is that mass absorption coefficient of silicon is higher for lower energies than that for higher energies, and hence, low-energy photons have a greater interaction probability with detector material.
|Figure 7: Energy response of cluster count algorithm at 50 mGy/h dose rate normalized to Cs-137|
Click here to view
It was observed that the pixel clusters obtained are brighter as well as smaller in diameter for lower energy field (few tens of keV) in comparison to the higher energy field, as shown in [Figure 8]. For higher energies, the longer lengths of clusters may cause clusters from two different events to overlap and be counted as one, but for lower energy photons, probability of overlap of two or more clusters is reduced. This phenomenon also contributes in getting higher than actual dose rate readings at lower energies. However, it has been observed that cluster count shows small increase for higher gamma energies in the range of few hundreds of keV.
|Figure 8: Pixel clusters observed in images captured in radiation field (a) 30 kev X-ray at 60 mGy/h (b) Cs-137 at 60 mGy/h|
Click here to view
| Conclusions|| |
A cluster count algorithm was developed for CMOS camera available in cell phones and tablets for its use as a radiation detector. The developed cluster count algorithm was tested to measure gamma dose rates in the range 5 mGy/h to 10 Gy/h. For dose rates <5 mGy/h, the cluster count algorithm shows poor sensitivity and is not suitable. A combination of cluster count algorithm and high-delta algorithm could be implemented for covering dose rate measurement range from μGy/h to Gy/h. The cluster count algorithm is found to be sensitive to energy of incident X-ray/gamma radiation photons, and suitable energy correction technique can be developed to get flat energy response.
Authors would like to thank Dr. K. S. Pradeepkumar, AD, HS and EG, BARC, for his support and encouragement during the course of this work. Authors are also thankful to Mr. Sunil Singh and Mr. S. M. Tripathi, RSSD, BARC, for actively helping in irradiating the tablet device used for the experiment.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Nakamura J. Image Sensors and Signal Processing for Digital Still Cameras. Florida; USA: CRC Press, Taylor and Francis Group; 2006.
Godbeer A. MPhys Dissertation: Investigation of 4T CMOS Image Sensor Design and the Effects of Radiation Damage. University of Surrey, Guildford, Surrey, GU2 7XH, UK; 12 February, 2010.
Drukier GA, Rubenstein EP, Solomon PR, Wojtowicz AA, Serio MA. Low Cost, Pervasive Detection of Radiation Threats. Technologies for Homeland Security (HST), IEEE International Conference; 2011.
Cogliati JJ, Derr KW, Wharton J. Using CMOS Sensors in a Cell Phone for Gamma Detection and Classification. arXiv:1401.0766v1 [physics.ins-det]. 2014.
Tian H. PhD Thesis: Noise Analysis in CMOS Image Sensors, Stanford University. 2000.
Gonzalez RC, Woods RE. Digital Image Processing. 3rd
ed. London, England: Pearson Education International; 2008.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6], [Figure 7], [Figure 8]
[Table 1], [Table 2]