INTRODUCTION TO IMAGE PROCESSING - Drkmm

Readings in Image ProcessingOVERVIEW OF IMAGE PROCESSINGK.M.M. Rao*,Deputy Director,NRSA,Hyderabad-500 037IntroductionImage Processing is a technique to enhanceraw images received from cameras/sensorsplaced on satellites, space probes andaircrafts or pictures taken in normal day-today life for various applications.Methods of Image ProcessingThere are two methods available in ImageProcessing.Analog Image ProcessingVarious techniques have been developed inImage Processing during the last four to fivedecades.Most of the techniques aredeveloped for enhancing images obtainedfrom unmanned spacecrafts, space probesand military reconnaissance flights. ImageProcessing systems are becoming populardue to easy availability of powerful personnelcomputers, large size memory devices,graphics softwares etc.Image Processing isapplications such as: usedinAnalog Image Processing refers to thealteration of image through electrical means.The most common example is the televisionimage.The television signal is a voltage level whichvaries in amplitude to represent brightnessthrough the image. By electrically varying thesignal, the displayed image appearance isaltered. The brightness and contrast controlson a TV set serve to adjust the amplitude andreference of the video signal, resulting in thebrightening, darkening and alteration of thebrightness range of the displayed image.variousRemote SensingMedical ImagingNon-destructive EvaluationForensic StudiesTextilesMaterial Science.MilitaryFilm industryDocument processingGraphic artsPrinting IndustryDigital Image ProcessingIn this case, digital computers are used toprocess the image. The image will beconverted to digital form using a scanner –digitizer [6] (as shown in Figure 1) and thenprocess it. It is defined as the subjectingnumerical representations of objects to aseries of operations in order to obtain adesired result. It starts with one image andproduces a modified version of the same. It istherefore a process that takes an image intoanother.The common steps in image processing areimage scanning, storing, enhancing andinterpretation. The schematic diagram ofimage scanner-digitizer diagram is shown infigure 1.The term digital image processing generallyrefers to processing of a two-dimensionalpicture by a digital computer [7,11]. In abroader context, it implies digital processing ofany two-dimensional data. A digital image isan array of real numbers represented by afinite number of bits.The principle advantage of Digital ImageProcessing methods is its versatility,repeatability and the preservation of originaldata precision.Figure 1* Deputy Director, National Remote Sensing Agency,Hyderabad, India. e-mail: dddp@nrsa.gov.in1

Readings in Image ProcessingThe various Image Processing techniquesare: Image PreprocessingScalingImage representationImage preprocessingImage enhancementImage restorationImage analysisImage reconstructionImage data compressionThe theme of the technique of magnification isto have a closer view by magnifying orzooming the interested part in the imagery.By reduction, we can bring the unmanageablesize of data to a manageable limit. Forresampling an image Nearest Neighborhood,Linear, or cubic convolution techniques [5] areused.Image RepresentationAn image defined in the "real world" isconsidered to be a function of two realvariables, for example, f(x,y) with f as theamplitude (e.g. brightness) of the image at thereal coordinate position (x,y). The effect ofdigitization is shown in Figure 2.I. MagnificationThis is usually done to improve the scale ofdisplay for visual interpretation or sometimesto match the scale of one image to another.To magnify an image by a factor of 2, eachpixel of the original image is replaced by ablock of 2x2 pixels, all with the samebrightness value as the original pixel.Figure 2The 2D continuous image f(x,y) is dividedinto N rows and M columns. Theintersection of a row and a column iscalled as pixel. The value assigned to theinteger coordinates [m,n] with {m 0,1,2,.,M-1} and {n 0,1,2,.,N-1} is f[m,n]. Infact, in most cases f(x,y)--which we mightconsider to be the physical signal thatimpinges on the face of a sensor.Typically an image file such as BMP,JPEG, TIFF etc., has some header andpicture information. A header usuallyincludes details likeformat identifier(typically first information), resolution,number of bits/pixel, compression type,etc.Figure 3 Image MagnificationII. ReductionTo reduce a digital image to the original data,every mth row and mth column of the originalimagery is selected and displayed. Anotherway of accomplishing the same is by takingthe average in 'm x m' block and displayingthis average after proper rounding of theresultant value.2

Readings in Image ProcessingMosaicMosaic is a process of combining twoor more images to form a single large imagewithout radiometric imbalance. Mosaic isrequired to get the synoptic view of the entirearea, otherwise capture as small images.Figure 4Image ReductionRotationRotation is used in image mosaic, imageregistration etc. One of the techniques ofrotation is 3-pass shear rotation, whererotation matrix can be decomposed into threeseparable matrices.3-pass shear rotationR cosα sinα 1 –tanα/2 01 –sinα cosα 1 sinα0 1 1 –tanα/2 01 Figure 6 Image MosaickingImage Enhancement TechniquesFigure 5Some times images obtained from satellitesand conventional and digital cameras lack incontrast and brightness because of thelimitations of imaging sub systems andillumination conditions while capturing image.Images may have different types of noise. Inimage enhancement, the goal is to accentuatecertain image features for subsequentanalysis or for image display[1,2]. Examplesinclude contrast and edge enhancement,pseudo-coloring, noise filtering, sharpening,and magnifying. Image enhancement is usefulin feature extraction, image analysis and animage display. The enhancement processitself does not increase the inherentinformation content in the data. It cs. Enhancement algorithms ass Shear RotationAdvantages1. No scaling – no associated resamplingdegradations.2. Shear can be implemented veryefficiently.3

Readings in Image ProcessingSome of the enhancement techniques are: Contrast StretchingNoise FilteringHistogram modificationContrast Stretching:Some images (eg. over water bodies, deserts,dense forests, snow, clouds and under hazyconditions over heterogeneous regions) arehomogeneous i.e., they do not have muchchange in their levels. In terms of histogramrepresentation, they are characterized as theoccurrence of very narrow peaks. Thehomogeneity can also be due to the incorrectillumination of the scene.Figure 8Ultimately the images hence obtained are noteasily interpretable due to poor humanperceptibility. This is because there existsonly a narrow range of gray-levels in theimage having provision for wider range ofgray-levels. The contrast stretching methodsare designed exclusively for frequentlyencountered situations. Different stretchingtechniques have been developed to stretchthe narrow range to the whole of the availabledynamic range.Figure 9Noise RemovalEdge EnhancemenHistogram ModificationHistogram has a lot of importance in imageenhancement. It reflects the characteristics ofimage. By modifying the histogram, imagecharacteristics can be modified. One suchexample is Histogram Equalization.Histogram equalization is a nonlinear stretchthat redistributes pixel values so that there isapproximately the same number of pixels witheach value within a range. The resultapproximates a flat histogram. Therefore,contrast is increased at the peaks andlessened at the tails.Figure 7 contrast stretchingNoise FilteringNoise filtering is used to filter the unnecessaryinformation from an image. It is also used toremove various types of noises from theimages. Mostly this feature is interactive.Various filters like low pass, high pass, mean,median etc., are available.Figure 10 Histogram equalized output4

Readings in Image Processingare in good registration. Most of theinformation extraction techniques rely onanalysis of the spectral reflectance propertiesof such imagery and employ specialalgorithms designed to perform various typesof 'spectral analysis'. The process ofmultispectral classification can be performedusing either of the two methods: Supervisedor Unsupervised.Image AnalysisImage analysis is concerned with makingquantitative measurements from an image toproduce a description of it [8]. In the simplestform, this task could be reading a label on agrocery item, sorting different parts on anassembly line, or measuring the size andorientation of blood cells in a medical image.More advanced image analysis systemsmeasure quantitative information and use it tomake a sophisticated decision, such ascontrolling the arm of a robot to move anobject after identifying it or navigating anaircraft with the aid of images acquired alongits trajectory.In Supervised classification, the identity andlocation of some of the land cover types suchas urban, wetland, forest etc., are known aspriori through a combination of field works andtoposheets. The analyst attempts to locatespecific sites in the remotely sensed data thatrepresents homogeneous examples of theseland cover types. These areas are commonlyreferred as TRAINING SITES because thespectral characteristics of these known areasare used to 'train' the classification algorithmfor eventual land cover mapping of rs are calculated for each trainingsite. Every pixel both within and outside thesetraining sites is then evaluated and assignedto a class of which it has the highest likelihoodof being a member.Image analysis techniques require extractionof certain features that aid in the identificationof the object. Segmentation techniques areused to isolate the desired object from thescene so that measurements can be made onit subsequently. Quantitative measurementsof object features allow classification anddescription of the image.Image SegmentationImage segmentation is the process thatsubdivides an image into its constituent partsor objects. The level to which this subdivisionis carried out depends on the problem beingsolved, i.e., the segmentation should stopwhen the objects of interest in an applicationhave been isolated e.g., in autonomous air-toground target acquisition, suppose ourinterest lies in identifying vehicles on a road,the first step is to segment the road from theimage and then to segment the contents ofthe road down to potential vehicles. Imagethresholding techniques are used for imagesegmentation.Figure 11. Image ClassificationIn an Unsupervised classification, theidentities of land cover types has to bespecified as classes within a scene are notgenerally known as priori because groundtruth is lacking or surface features within thescene are not well defined. The computer isrequired to group pixel data into ly determined criteria.ClassificationClassification is the labeling of a pixel or agroup of pixels based on its grey value[9,10].Classification is one of the most often usedmethods of information extraction. InClassification, usually multiple features areused for a set of pixels i.e., many images of aparticular object are needed. In RemoteSensing area, this procedure assumes thatthe imagery of a specific geographic area iscollected in multiple regions of theelectromagnetic spectrum and that the imagesThe comparison in medical area is thelabeling of cells based on their shape, size,color and texture, which act as features. Thismethod is also useful for MRI images.5