7 pages

A knowledge-based stellar image interpretation system

of 7
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
A knowledge based system for interpreting astronomical digital images has been developed. The system examines a given stellar image produced by special light sensitive detectors, and employs rule-based knowledge about the light intensity profile of
    Al Azhar University Engineering Journal, Vol. 1, No. 2, pp. 114-120, 1998Published by Faculty of Engineering, Al Azhar University, Cairo - Egypt    ISSN 1110  –  6409  1998 AUEJ All rights reserved A KNOWLEDGE-BASED STELLAR IMAGE INTERPRETATION SYSTEM Mohamed A. Madkour*, A. El-Bassuny Alawy**, M. S. Ella** and Farag I. Younis** * Systems & Computers Engineering Dept., Faculty of Engineering, Al-Azhar Univ.** National Research Institute of Astronomy and Geophysics, Helwan. Abstract A knowledge based system for interpreting astronomical digital images has been developed. The system examines a given stellarimage produced by special light sensitive detectors, and employs rule-based knowledge about the light intensity profile of typicalstars in order to identify the true stars and filter out other objects. Forward chaining is used initially to scan the light intensity in agiven image searching for pixels having local maxima. Next backward chaining is employed to classify the obtained maxima intotrue stars, cosmic rays, and noise peaks. The developed system is both simple and fast. It is implemented in C language and runson a personal computer. Obtained results are found to be in good agreement with other obtained from more complex traditionalpackage. Keywords : Knowledge-based system, Stellar images, Image processing. 1. INTRODUCTION Astronomical optical telescopes are considered thebasic instruments for observing the celestial objects. Soastronomers built new large modern telescopes ordeveloped their imaging systems to enhance their dataacquisition ability. The higher quantum efficiency of theastronomical imaging system, the maximum the abilityof the system to observe very faint objects in short timeand the better the accuracy of the image obtained, evenwith moderate size telescopes. The advent of ChargeCoupled Devices (CCD) imaging has spawned averitable revolution in astronomy. In truth, the greatestlimitation of astronomy has always been, and still is, thefact that faint objects are being imaged [1,2].CCD detectors are capable to record a huge numberof star images on a single frame. When the number of stellar images in a data frame is small, the task of identifying each image is easily performed by a humaneye and brain examining a pictorial representation of adigital image. As the number of stellar images ismoderately increased doing this task by hand and eyewould be an atrocious waste of human effort.Furthermore, the CCD can record data over a muchwider dynamic range of brightness than the human eyecan perceive, all of which can be exploited by anautomatic star interpretation system [3].At any given picture element (pixel) in the CCDimage, the data number supposedly representingdetected photons come from any of several sources : (1)detected stars, (2) undetected stars, (3) localized imagedefects and cosmic ray events, and (4) diffuse sources,including but not necessarily limited to (a) the terrestrialnight sky, (b) scattered light in the camera, and (c) someother astronomical sources [4,5].For greatest usefulness the star interpretation systemshould have the ability to distinguish brightnessenhancements which correspond to actual stars,however blended from the other sources cited above.In the last two decades, sophisticated detectors forlow light level have been designed and upgraded. In  Mohamed A. Madkour et al. : A Knowledge-Based Image Interpretation SystemAl Azahar University Engineering Journal, Vol. 1,No. 2, July 1998      addition, personal computers have been introduced anddeveloped as well as powerful workstations. In view of these advents, astronomers have been able to constructvarious software to handle stellar image frames,employing numerical fitting techniques based on pre-assumed mathematical approximations [5]. One of themost common such packages is DAOPHOT [4,5],which will be considered in the present work as areference for comparison.This paper presents the development of anintelligent system for stellar image interpretation, and isstructured as follows. Section 2 specifies the problemstatement and objective. In section 3, the developedsystem is generally described while its implementationdetails are given in section 4. A case study is selectedand used to verify the system applicability in section 5where a comparison is presented between the obtainedresults and the published ones. Section 6 deals with thepractical considerations of the present system. Someconcluding remarks concerning the present work aregiven in section 7. 2. PROBLEM STATEMENT AND OBJECTIVE The main objective of the present work is to developa simple and fast stellar images interpretation system(SIIS) that should be able to identify stellar identities inCCD image frames. The major problem facing this task is that the obtained images contain many identities,some of which are actually due to stars, while the othersare not though their brightness distribution are almostas that of stellar srcin . The latter may be due to somesources as stated in section 1. Hence in order to identifya stellar image properly one must design a reliableapproach having the ability to differentiate between truestars and other identities. The approach deemed here isbased on knowledge-based systems. 3. GENERAL DESCRIPTION OF THEDEVELOPED SYSTEM KB-SIIS is a knowledge based system that examinesstellar images available in industry-standard formats [6]and employs heuristic rules to identify stellar identitiesin a given frame. The system accepts a CCD imageframe as an input and examines it to identify thelocations of true stars in the given image and filter outother noise effects. In essence, it acts as a classifier thatscans the given frame to locate spots of relatively highlight intensity and classifies each of them as either atrue star, cosmic ray, or noise peak. SIIS employssimple forward chaining [7] rules to obtain ,through thescanning process, a list of high intensity pixelsrepresenting local maxima. Next, each pixel in that listis compared to its neighbour pixels using heuristicbackward chaining rules in order to classify it into oneof the above mentioned three types. The classificationtakes place on two successive stages as will be shownnext. 4. DETAILS OF IMPLEMENTATION Fig. 1-a shows the two dimension light intensitydistribution in a star image. As shown in Fig, 1-b, thebasic star image profile can be modeled as follows[8]:1) The central part of the star image is a nearly uniformdisk which is surrounded by a region of steeply fallingbrightness. It is approximately Gaussian in form.2) The outer part of the profile is well represented by apower law.3) The In -  Between part of the profile is a transitionregion, which can be represented by an exponential lawor any similar function.(a) Light intensity in a two dimensional image. Central part   In-between part   Outer part  (b) Cross section model for a star image profileFig. 1 The star image.  Mohamed A. Madkour et al. : A Knowledge-Based Image Interpretation System   Al Azahar University Engineering Journal, Vol. 1,No. 2, July 1998    Thus, a star image is inherently a complex object.The physical srcin of these separate parts of the profileis not clear. Theories of atmospheric seeing exist butthey do not seem to predict unique mathematical shapeof any except the small Gaussian part of the curve. I(x-2,y-2) I(x-1,y-2) I(x,y-2) I(x+1,y-2) I(x+2,y-2)I(x-2,y-1) I(x-1,y-1) I(x,y-1) I(x+1,y-1) I(x+2,y-1)I(x-2,y) I(x-1,y) I(x,y) I(x+1,y) I(x+2,y)I(x-2,y+1) I(x-1,y+1) I(x,y+1) I(x+1,y+1) I(x+2,y+1)I(x-2,y+2) I(x-1,y+2) I(x,y+2) I(x+1,y+2) I(x+2,y+2) (a) 5x5 window.(b) Directions for checking the intensity change.Fig. 2 Schematic diagram of the pixel array intensitydistribution.The developed system uses the features of thecentral part of the star image, to identify stars in theCCD image frame. We code this in the knowledge baseusing a set of rules that infer this brightness distribution.The system scans the entire image, looking for a regionhaving brightness higher than its surrounding. If suchregion was found, the system presumes the presence of a star and the coordinates of such pixel represent thestar's center.The following simple commonsense logic is used toidentify the local maxima. Let x and y represent thecoordinates of the central pixel whose number of electrons collected are parametrized by I(x, y).Referring to Fig. 2-a, the light intensity of the othercontiguous horizontal pixels will be denoted by I(x-1,y),I(x+1,y),...etc.. Similar representations is followed forother neighbouring pixels..An important star feature can be determined uponexamining the characteristics of image brightnessdistribution shown in Fig. 1-a. That is the center of thedistribution has the peak value and the values decreaseoutward in all directions. In the present work, weconsider a 5x5 window with the brightest pixel in itscenter as shown in Fig. 2-a. The following inferencerules are designed to check the above-mentioned starfeature. It should be noted that a 5x5 window issufficient to show the direction of change in brightness,and there is no more use in adopting a larger sizewindow.An extremely important feature used alwaysto discriminate between stars and cosmic rays is thesharpness of the light intensity profile. Actually, acosmic ray event will be much sharper compared to atrue star. Referring to the 5x5 window of Fig. 2-a, a metric can be assigned as the “SHARPNESS” val uewhich is the ratio of the light intensity at the centralpixel to the sum of intensities at the 8-neighhbouringpixels. It represents the sharpness of the light intensityprofile. This metric will be used in the rule base, andis given by [4]: sharpness       I x y I x i, y j  ji ( , ) ( ) 11111 (1) [( i    0 and j    0 )]   In the following heuristic rules, the first one checksfor the presence of a peak intensity pixel. Given suchpixel, the other rules check for intensity changes in alldirections as shown in Fig. 2-b.     Rule to find a local maximum intensity pixel  IF I( x, y ) > I( x-1, y-1 ) AND I( x, y ) > I( x-1, y ) AND I( x, y ) > I( x-1, y+1 ) AND I( x, y ) > I( x, y-1 ) AND I( x, y ) > I( x, y+1 ) AND I( x, y ) > I( x+1, y-1 ) AND I( x, y ) > I( x+1, y ) AND I( x, y ) > I( x+1, y+1 ) THEN Central pixel is a local maximum     Rule to check the intensity decrease    along theU-direction IF Central pixel is a local maximum AND I( x, y-1 ) > I( x-1, y-2 ) AND I( x, y-1 ) > I( x , y-2 ) AND I( x, y-1 ) > I( x+1, y-2 ) AND I( x, y-1 ) > I( x-1, y-1 ) AND I( x, y-1 ) > I( x+1, y-1 ) THEN The intensity decreases in the U direction. Similarly, there are three other rules to check theintensity decrease in the L, R, and D directions.     Rule to check the intensity decrease    along the   UL-direction . IF Central pixel is a local maximum AND I( x-1, y-1 ) > I( x-2, y-2 ) AND I( x-1, y-1 ) > I( x-1, y-2 ) AND I( x-1, y-1 ) > I( x , y-2 ) AND I( x-1, y-1 ) > I( x-2, y-1 )  Mohamed A. Madkour et al. : A Knowledge-Based Image Interpretation SystemAl Azahar University Engineering Journal, Vol. 1,No. 2, July 1998      AND I( x-1, y-1 ) > I( x-2, y ) THEN The intensity decreases in the UL direction. Again, there are three similar rules to check theintensity decrease in the UR, DL, and DR directions.    True star rule IF Central pixel is a local maximum. AND The intensity decreased in the U direction. AND The intensity decreased in the L direction. AND The intensity decreased in the R direction. AND The intensity decreased in the D direction. AND The intensity decreased in the UL direction. AND The intensity decreased in the UR direction. AND The intensity decreased in the DL direction. AND The intensity decreased in the DR direction. THEN TRUE STAR is centered at the pixel (x,y) The last rule scans the working memory looking formatches to its premises, if matches are found, then aTrueStar centered at pixel (x,y) is assigned. By applyingthe stated rules to several stars in several standard CCDastronomical images, the following two cases have beenobserved : Case 1 : The central part of some stars satisfies allprevious rules. Hence, stellar images can be directlyidentified without ambiguity.   Case 2 : The central part of some stars partially satisfiesthe previous rules. Such situation can be due to: a) lowsignal to noise ratio for faint stars, b)the light of blendedstars affect each other, c) improper preprocessingtechniques for CCD image frame which causes thecentral part of some stars to deviate from the previousrules. However, a bright pixel can be classified initiallyas a MayBeStar if it satisfies either one of the followingconditions.Referring to Fig. 2-b, the light intensity should :a) decrease in at least two directions from the {U, L, D,and R} set, , and at least in two directions from the{UL, UR, DL, and DR} set., or b) decrease in at least three directions from the {U, L,D, R} set.Consequently, further processing should be carriedout to determine the nature of a MayBeStar pixel. Forexample, the following rule is used to check for theconditions of case (2-a). MayBeStar centered at the pixel (x,y) IF The intensity decreases in the U direction, AND The intensity decreases in the L direction, AND The intensity decreases in the UL direction, AND  The intensity decreases in the UR direction. Similar rules can be developed based on theconditions stated in case 2 above. However, uponapplying such rules, the resulting set of MayBeStarpixels would include other stellar identities in additionto true stars. Namely, these are cosmic rays and noisepeaks.Fig. 3 shows the phases used in the developed SIISsystem to identify and classify the different entities in aCCD frame. The system scans the entire frame pixel bypixel using forward chaining inference technique,looking for regions whose brightness distribution obeysthe conditions stated in either case 1 or case 2. If suchregions was found, the system produces two lists.i) A TrueStar list for all pixels that fulfill theconditions of case 1.ii) A MayBeStar list for all pixels obeying case 2.For each identity in both lists, the followinginformation are recorded :    the coordinates of the brightest pixel at the centerof the region.    the intensity of that pixel.    the sharpness value computed by equation 1. StartRead image frameForward chaining process to identify the pixelshaving local maxima (of light intensity) andclassify them as either TrueStar or MayBeStarDetermine minimum intensity andmaximum sharp from the TrueStar listBackward chaining process to examinethe MayBeStar listin order to classify its contents into :TRUE STAR, COSMIC RAY, and NOISEStop Fig. 3 Classification phases in the KB-SIIS system.  Mohamed A. Madkour et al. : A Knowledge-Based Image Interpretation System   Al Azahar University Engineering Journal, Vol. 1,No. 2, July 1998    Through these phases two parameters are evaluatedfrom the TrueStar list which are:    Threshold value : which is the minimumintensity value.    Maximum Sharpness : which is the maximumsharpness value.The KB-SIIS system automatically computes theseparameters for each entry in the MayBeStar list to allowobjective classification without user intervention.However, a user interaction mode is provided in theSIIS system in order to allow an experienced user toenter a better threshold value which may lead to find outmore true stars.Three backward chaining rules are then used toclassify the entities of the MayBeStar list as truestar,cosmic ray, or noise peak. These rules are :  A MayBeStar centered at pixel(x,y) is a TRUE STAR  IF   The intensity at (x,y)  Threshold value, AND  The sharpness at (x,y)  Maximum Sharpness.  A MayBeStar centered at pixel(x,y) is a COSMIC RAY   IF  The intensity of that pixel  Threshold value, AND  The sharpness at (x,y) > Maximum Sharpness.  A MayBeStar centered at pixel(x,y) is NOISE  IF   The intensity of that pixel < Threshold value. 5. CASE STUDY Fig. 4 shows a picture of the test case which is aCCD image of the star cluster M67 [9]. It ischaracterized by:    Size of 320 x350 pixels,    Optical filter used is visual filter,    Exposure time =30 Sec.    Maximum pixel value = 16252 ADU (Analog /Digital converter unit),    Minimum pixel value = 9 ADU.Upon analysing this image using the DAOPHOTpackage, 134 stars have been reported via the userinteraction mode where the threshold value is providedby the user. On the other hand, using the same thresholdvalue, the developed KB-SIIS system has identified 131stars, two of which are not identified by DAOPHOT.This shows that the results obtained by both systems arein a very good agreement, which is clearly depicted inFig. 5. Detailed astronomical discussion andinterpretation are being [10]. 6- PRACTICAL CONSIDERATIONS While our main purpose is primarily concerned topropose a KB-system for identifying stellar images onCCD frames, some other useful applications werefound. The developed new approach provides anefficient and quick tool to evaluate the quality of individual CCD images as well as the whole frame. Thiscan be evident through the following aspects.Fig. 4 CCD image of the star cluster M67      Adopting the proper exposure time (ET) : After obtaining a trial frame in an observing night,the SIIS system can be used to find pixels havingmaximum and minimum electron number,corresponding to the brightest and faintest starsobserved respectively. If the Max value found (or that of the target object) is less than the Full Well Capacity(FWC) of the chip used, then it is advisable to increasethe ET to enhance signal-to-noise (S/N) ratio. On theother hand, when some contiguous pixels (or one pixel)were found to have electron number close ,or equal tothe FWC, one concludes that the ET used is too longand has to be shortened. Since the SIIS system takesfew seconds for a medium size frame it serves an as online tool to adopt the proper ET for the whole frame orstars of interest.     Check the optical alignment: It is necessary to set the CCD chip surfaceperpendicular to the telescope principal optical axis.Any deviation (tilt or shift) from such position willproduce elongated images for stars being observed.When snapping an over-exposed frame for a field of 
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!