MORPHOLOGICAL IN 4, Erosion a) sharps a region c) increases a region i Answer: (c) J. ceeessnneerees os is not a morphological image - a) Thinning c) Both (a) & (b) Answer: (d) 3,. Knowledge of which one of the follov image processing? a) Erosion c) Neural networking ee fAYNS PULA Illustration SS NSS \ WOL Yj Y LF. N YY Yj WW 3 4 5 6 oo YWUGG NSN NW WC ie NSW - 70 ANN A: Set AThe original of the pixel is included in the c uite possible because for certain types of r; reduction capabilities with considerably less size. These are effective for bi-polar and unipc 3. a) Briefly explain skeletons. b) Write down the effects of the dilation an Answer: a) Skeleton are the minimal representations enter number of the image. The enter number the number of holes in those objects. The skel by successive thining until stability is reached ’b) Effect of dilation Process: |. Dilation expands the input image. 2. This process effects both the inside and out: 3. Dilation fills the holes in the image. 4. The dilation Process smoothens-out the ima Ce ‘+ = a ,POPULAR PUBLICATIONS is a property attached to an individual pixe behavior in a neighbourhood of the pixel. | presence of the edge. Sign of the second deri on the dark sign or lightside. , Pre hori: deri Sec deri’ Bi Fig: (a) Fig. Edge detection by derivative operators. a) light stripe on a dark background; segments connecting every pair of pixels. Thi principal reasons: 1) The resulting chain tends to be quite | 2) Any small disturbances along the segmentation cause changes in the cc shape features of the boundary. ' An 2pproach frequently used to circumvent t by selecting a larger grid spacing as shown bePUBLICATIONS The resampled boundary obtained in this below shows the coarser boundary points a simple matter to convert from an 8-code The starting point in the on is 3 ne gives the chain code 0766 . code representation depends « on ‘a s ~~ |Answer: a) We can normalize for rotation by using the f the code itself. The difference is obtained by « . (in counter clockwise direction) that separate example, the first difference of the 4-direction elect to treat the code as a circular sequence, computed by using the transition between the I The result is 33133030. b) 0110233210332322111 The first difference will be 103210333303130300 by using above diagram. c)POPULAR PUBLICATIONS Optical Character recognition Pattern matching Face recognition License plate matching Scene change detection Pattern Matching: A Pattern is an arrangement of descriptors: recognition literature to denote a descriptor share some common properties. Patterns cl the number of classes. Pattern recognition | patterns to their respective classes automati possible. Three common patterns are-vector: 4. Write short notes on the following: a) Morphology : b) Hit-or-miss Transform c) Erosion and Dilation d) Chain code Answer: ZS a) Morpholoeyv: ~the image and then comparing the structur in ies grosion and Dilation: jon and Erosion are two fundamental mort pilatt poundaries of objects in an image, » | “ ries. The number of pixels added to ynaaris>- | ds on the size and shape of the structuring depen Chain code: Refer to Question No. 1 of I )POPULAR PUBLICATIONS IMAGE SEGMENTATI AND DE! | Multiple Choi 13 Representation & description almost a) Segmentation stage c) Compression stage Answer: (a) 2. Isolated point can be best detected l a) 1°" derivative c) 3° derivative Answer: (b) 3. What operation is needed for barry a) Ilmaae multiplicationpt The region are grown from: these seed poi st hold oF criteria (8-connected) we make. { 3 R egion growth should stop when no more F : region. Here suitable selection of seed point in image iS better. The result of an image after ri a higher than the threshold but not connected wit jwite down the usefulness of segmentation? wer: oe a nage segmentation ts typically used to locate obje: images: More precisely image segmentation is th ‘ye! in an image such that pixel with the same lat entation is the process of dividing an image al to identify objects or other relevant informatio 1, consider the image segment 3 | 2 2 2 0 1 2 4) / a oePOPULAR PUBLICATIONS | Long Answe 1. a) What is boundary descriptor? b) Explain Fourier descriptor. Answer: a) There are several simple geometric - boundary, The length of a boundary w gives a rough approximation of its lengtl Curvature is the rate of change of slope. digital boundary is difficult. The diffe segments is used as a descriptors of curv Order 4 Chain code 0321 0032: a inet Se aan: an articFourier _— iptors are a way of encoding th raking the Fourier transform of the boundary, wh is mapped to 8 complex number x+iy. The or inverse Fourier transform. pet S(k) be a co-ordinate of a boundary point k | S(k) = x(k) +iv(k) fourier descriptor: a(u)=7>,S(k)e Fourier descriptor of a shape vaginary axis > . ,e@e@e8e8 eo @POPULAR PUDLICATIOND 2. What do you mean by global a difforence botwoon raglon growing an Auawer: I“ Part: In peneral, the threshold can be chron P(r, c) is the feature value at the pi P(ree)> g(r, c) for all (r, ¢) within p(r. c) only, it is called local threshol well as on the feature P(r, c) at that pir 2" Part: Region-growing, methods rely mainly ¢ within one region have simjlar values, ’ with its neighbours, a similarity criter the cluster as one or more of ils neigh partition of an i image. I is sometimes cal the root of the tree that repres cots AOare three basic ways to threshold a tras ihe id can be applied to all subimages, (2) A ines age, (3) The threshold can be varied as ‘aC ent within the subimage. ¢ Minimum mean-square error estimation | roration of a particular image. This may be o ance of any knowledge about noise we ma pane a particular image, we minimize the dif w re-degraded estimated image. a method of constrained least square can | efterion of goodness or quality of image sub igtween the image and the re-degraded estimate yect0F. Constraint least square is a linear ima gstoration is not. ; Constraint least square retain the simplicity of a prior knowledge of mean and variance of the no When point-spread function is symmetric, properly, to select a solution a constraint le ,POPULAR PUBLICATIONS _ system should assign correct labels to re object recognition problem is closely tied partial recognition of objects segmentati object recognition is not possible. Different components of an object recog An object recognition system must have t! © Model database (also called mode e Feature detector ¢ Hypothesizer ¢ Hypothesis verifier The model database contains all the mod model database’ depends | on the approac S80 ce lg° if only If false. split the area of interest (usua each of the sub-area as the areq of inte This process Continues until no furtt happened when the areas are just one | This is a divide-and-conquer or top- a splitting schedule is used then the. any neighbouring regions that have identica hus, 3 merging process is used after each merges then if necessary, algorithms of this n; To iJlustrate the basic principle of these meth¢ Let | denote the whole image shown i Not all the pixels in I are similar so tk Assume that all pixels within regior those in /, are not. Therefore /, is split next as in Fig. (i Now assume that all pixels within \ region and that after comparing the:POPU! PUBLICATIONS The point is detected at a location (x. y) ir corresponding value of R such that | R | > Where R is the response of the mask at a threshold value. It means that isolated poin This formulation serves to measures the ¥ and Its neighbours since a. grav level of an of its neighbors. Line detection is the next level of complex any point in the image, a Tesponse can be point of a line is most associated with. Fo: mask. Then we have. IR] >[ ab v, =i It means that the corresponding points is r direction of the mask 7. Line detection direction (fig. (b)). vertical direction (fig. 9) 12 ate2it is however for more efficient to use the ima yeral methods for finding the “optimal” t threshold provides the following methods for ¢ l. 2. Automatically calculate a threshold va ‘Approximately the histogram of the it a mid-point value as the threshold leve _ Fuzzy thresholding using as the meas Evaluate the threshold based on the rows (A@aptive thresholding), b) Constrained least squares: The constrained least square filter is anoth difficulties of the inverse filter (excessive n (estimation of the power spectrum of the ide< _ of a spatially invariant linear filter. But it is mean and variance of the noise. The constrai * adirect solution using a criterion C , which ¢ image. It is a linear image restoration techni image is maximized subject to a constraintorPOPULAR PUBLICATIONS algorithm starts from the initial assumption computes the homogeneity criterion to sée if split into the four smaller régions. This pr regions until no further splitting is necessary if they are similar to give larger irregular reg merges are possible.UESTIC . GROU Oiseye Choice 4, choose the correct alternatives for any ten of tt i) Colour image processing is gaining importance a) It's more pleasant to watch b) It's cost effective c) It's easy to capture and represent ¥ 4d) Use of digital image over the internet ha a) Wavelets are ; ¥ a) Foundation for representing image in vz b) Of remarkabie advantage for i image come c) Both (a) and (b) d) None of the above iii) Representation & description almost always | Y a) Segmentation stage . a ,"See Topic: DIGITAL IMAGE PROCESSIN ix) Which OF iy Ya) Block transfer coding _¢) LZW coding x) Isolated point can be best detected using a) 1* derivative c) 3% derivative xi) In the intensity distribution scale the backgrou %a) Lower intensity value c) Medium intensity value: xii) Measuring an intensity value of a fixed pixel i a) Averaging ., Yc) (a) & (b) both “ GRO (Short Answer 2. What is Weber ‘Ratio? Show the variation of V See Topic: DIGITAL IMAGE PROCESSING SYS 3. What are the basic steps involved in < homogeneous from of this transformation? .” +? ; WY ~ 4A al lAsfnat AR wri maanm bk...) ) & b) See Topic: IMAGE ENHA a NCEMENT 1 short Answer Type Question No. 7, ) See Topic: IMAGE TRANSFORMS (IMPLEME : ¢ 8. a) Explain the operation of a Median filter. b) Describe Bilinear interpolation method. What is Hough transform and wher. a See Topic: IMAGE ENHANCE panel Type Question No. 8, cae Topic: DIGITAL IMAGE PROC ¢) See Topic: IMAGE TRANSFORMS ( € is it used? MENT IN TH ESSING s IMPLEME ga) Write down the usefulness of 5 b) Briefly explain about region growing technique ¢) Explain the role of discrete cosine transform in a) Sce Topic: IMAGE SEGMENTATION, REPRE Type Question No. 2. ; b) See Topic: IMAGE SEGMENT ATION, REPRE Type Question No. I(or). . % ¢) See Topic: IMAGE TRANSFORMS CIMPYEMI €gmentation’ .POPULAR PUBLICATIONS g) Split and Merge Algorithm. a) See Topic: IMAGE ENHANCEMENT IN THE § Answer Type Question No. 11(e). b) See Topic: IMAGE SEGMENTATION, REPRESE Type Question No. 6(d). c) See Topic: DIGITAL IMAGE PROCESSING SYS’ d) Sce Topic: IMAGE TRANSFORMS (IMPLEMEN ¢) See Topic: MORPHOLOGICAL IMAGE PROCE: f) See Topic: IMAGE ENHANCEMENT IN THE | Answer Type Question No. 1 1(a). g) See Topic: IMAGE SEGMENTATION, REPRES Type Question No. 6(e). UESTIC GROU! (Multiple Choice ” 1. Answer any fen questions: ; i) The computation of Walsh coefficient involvvil) The smallest units Of a digital image is represe a) Aone dimensional matrix c) Dot . viii) Which of the following statement is true? a) The resolution of CMOS image sensor is t ¥b) The resolution of CCD image sensor is t c) CCD image sensor is cost effective than C d) none of the above . ix) What operation is needed for background rem: a) Image multiplication c) Image subtraction x) Wavelets are a) of remarkable advantage for image compr ¥b) foundation for representing images in vz c) both (a) and (b) d) none of these % xi) An example of volume image is SS a) A one dimencinnal imana icLAR PUBLICATIO} 5. Explain the operation of an ord . er statistics fi See Topic: MORPHOLOGICAL I iter ar MAGE PROCESSI 6. Discuss briefly about the usefulness of discrete ct See Topic: IMAGE TRANSFORMS (IMPLEMENTA GROUP (Long Answer Ty 7. a) What is Histogram of an- image and why specification? . . : by What is the difference between spatial domain fi . €dge detection operator and contrast stretching of c) Briefly describe the smoothing linear spatial filter a) See Topic: IMAGE TRANSFORMS (IMPLEMED b) & c) See Topic: IMAGE ENHANCEMENT IN Long Answer Type Question No. 7. 8. a) What is m-connectivity among pixels? Give a b) What do you mean by global and local thre: reaion crowing and split merae technique?11. Write short notes on any three of the following a) Local enhancement . b) Vector quantization method for image com Cc) Piecewise Linear transformation d) Erosion and Dilation a e) Image sharpening filters ~ f a) See Topic: IMAGE ENHANCEMENT IN THI Answer Typé Question No. 11(f. b) See Topic: IMAGE DATA COMPRESSION, Lo c) See Topic: IMAGE TRANSFORMS (IMPLEME d) See Topic: MORPHOLOGICAL IMAGE PROC ¢) See Topic: IMAGE ENHANCEMENT. §IN THI Answer Type Question No. 1 1(g). UESTI Grou (Multiple Choice 1. Choose the correct alternatives for i) The spectrum of the visible light is a) 10-350 nm any ten of tPOPULAR PUBLICATIONS vii) OCT is widely used for a) image degradation c) image restoration viii) The photosensitive detector of the human eye a) eye lens b) iris ix) Through decimated by 2 operations the sampli a) increased vb) decreased x) The histogram equalization process a) blurs the image Yc) improves the brightness of the image xi) If the size of the mask for averaging is increas a) noise frée Yb) blurred xii) Salt and pepper noise can be removed by a) weighted average filter ‘ “c) Median filter ‘ a, . _ * xiii) An example of dictionary based codina techn4, a) What do you understand by the terms ‘histog what are the steps to perform histogram equali . ‘What is the necessity for histogram equalizatior ) see Topic: IMAGE TRANSFORMS (IMPLEME p) & €) See Topic: IMAGE TRANSFORMS (IMPL 13(a) & (b). . ‘ " 4,a) What do you mean by aliasing in the context p) What do you mean by the term ‘image file for file format. . See Topic: DIGITAL IMAGE PROCESSING SYS 5, a) Briefly explain skeletons. b) Write down the. effects of the dilation and erosi sce Topic: MORPHOLOGICAL IMAGE PROCE 6. Show how Prewitt and Sobel operators can be See Topic: IMAGE ENHANCEMENT IN TH Answer Type Question No. 17. ® \GRPOPULAR PUBLICATIONS a) & c) See Topic: IMAGE SEGMENTATION, R Answer Type Question No. 4(b) & (c). b) See Topic: MORPHOLOGICAL IMAGE PROCE 10. a) What is the need for compression? What are b) Discuss briefly the concept of Huffman coding. Cc) Define interpixe! redundancy? See Topic: IMAGE DATA COMPRESSION, Long / 11. Write short notes on any three of the following: a) Walsh Transformation b) Image Restoration model C) Hotelling Transformation “d) Run length coding e) Chain code ' a) See Topic: IMAGE TRANSFORMS (IMPLEME! b) See Topic: IMAGE ENHANCEMENT IN THE Answer Type Question No. 11(h). c) See Topic: IMAGE TRANSFORMS (IMPLEME! d) See Topic: IMAGE DATA COMPRESSION, Lor ec) See Topic: MORPHOLOGICAL IMAGE PROC!‘ iv) The image function (x, y) is characterized S(%¥)=i(x, y).r(x,y) where a) O Ci TiTi the constant multipliers in front of each mas! yalues of its coefficients, as is required to comp st Part: . , pixels are digital number composed of bits. Fo 956-level gray-scale image is composed of 8 | ranges, we could highlight the contribution mi its. a illustration, an 8-bit image may be consi planes, with plane of containing the lowest orde all the highest order bits. One 8-bit byteM-1 - f(x)= DFA h=0 . for x=0,1,2,....,M-l. Taking ar complex = of Eqn. (2) at aS *( (x)= Lyre (ive Pe ial M i erties this result (3) with (1), it shows Fourier transform. Therefore, inputting F *(h) into an alg transform gives the quantity f*(x)/M .T by M — the desired inverse f(x). As M-1N-1 (x, =D MN i=0 v=0 which is in the form of a 2-D forward Fouri If f(x) or f(x, y) are real functions (e.g. left on Eqn. (4) is unnecessary; we simply complex term. _ 12. What is Quad-tree? : Answer: - The quad-tree data structure is widel un graphics for modelling spatial ok eSAnswer: 1" Part: . . salt and PepPer noise is also know nd sudden disturbances jn the im 7 black (or both) pixel over the j ie block diagram of salt & pep nas impul: age signal, Mage, Per noise Probability Salt & Peppe: 9" Part: ; . Gaussian noise is caused by random fluctuatic density function. It is useful for modeling nat Probability : & Probability —| harem en cs 22s 1 >POPULAR PUBLICATION Answer: The image restoration process model is where (i, j) is the degraded image JF (i, j) is the original image H is an operator that represents th n(i, j) is the external noise whic! Fig. below is a schematic diagram for a | above model. 9 Restoration method can be classified as de Since y(i, j) = H[f(i, j)]+nli,j) - If we ignore the presence of the external ni Wi D=HYGD) 9 His linear of = ~ HIk Sf (i,j) +k fi.) = HTTwhat is Holomorphic Filtering? Setup the 6, Answer ephic filtering uses illumination mom the image is been character; onsider is the amount of source illuminatic 7 second component is the reflectance i(% 7 The image is then defieg as 1207" , S98" ral logarithm of both sides to d Take @ | — reflec ized by ecoupl Z(x,y) =n (x,y) + fnr(x, yy siage? Fourier transform to transform the image uset ‘ S{Z(x, y)} = len i(x, y)}+ Slr or Z(u,v) = F, (u,v) + F, (u,v) re F(u,v) and F,(u,v) are the fourier tra whe me respectively. % aoe the z(u,v) by means of a filter functio Ig} a , =. @af.*