  ,((( ,PDJH )LOWUDWLRQ XVLQJ *UDSKEDVHG /RZ 3DVV )LOWHU 1DJDUDMX 6RQWL 'HSDUWPHQW RI (&( 9LJQDQV )RXQGDWLRQ IRU 6FLHQFH 7HFKQRORJ\  5HVHDUFK 9DGODPXGL $QGKUD 3UDGHVK ,QGLD QDJDUDMXVRQWL#JPDLOFRP 5XNPLQL 0 6 6 'HSDUWPHQW RI (&( 9LJQDQV )RXQGDWLRQ IRU 6FLHQFH 7HFKQRORJ\  5HVHDUFK 9DGODPXGL $QGKUD 3UDGHVK ,QGLD PVVUXNPLQL#JPDLOFRP 9HQNDWDSSD 5HGG\ 3DPXODSDWL 'HSDUWPHQW RI (&( 9LJQDQV )RXQGDWLRQ IRU 6FLHQFH 7HFKQRORJ\  5HVHDUFK 9DGODPXGL $QGKUD 3UDGHVK ,QGLD YHQNDWDSSDUHGG\S#JPDLOFRP $EVWUDFWｲ 7KLV SDSHU LV EDVHG RQ ILOWHULQJ YDULRXV QRLVHV IURP WKH LPDJH GDWD WKURXJK WKH JUDSKEDVHG ORZ SDVV ILOWHU 7KH OLQHDU QRLVH VDOW  SHSSHU QRLVH DQG QRQOLQHDU QRLVH *DXVVLDQ QRLVH DUH WDNHQ IRU WKH UHVHDUFK SXUSRVH WKRVH QRLVHV DUH UDQGRPO\ DGGHG WR WKH LPDJH GDWD ,Q WKLV SDSHU WKH JUDSK ILOWHU LV SURSRVHG WR UHPRYH WKH OLQHDU DV ZHOO DV QRQ OLQHDU QRLVH 7KLV JUDSKEDVHG ORZ SDVV ILOWHU DOVR DOORZV WKH QRLVH IURP WKH EXON LPDJH GDWDVHW ,Q WKH JUDSKEDVHG ORZ SDVV ILOWHU WKH LPDJH GDWDVHW LV FRQYHUWHG LQWR D VWUXFWXUH OLNH D JUDSK DQG WKHQ WKH LPDJH LV FRQVLGHUHG DV PDWUL[ GDWD 7KH SL[HOV LQ WKH GDWD FRQQHFWHG ZLWK QHLJKERULQJ SL[HOV WKDW GHPRQVWUDWH WKH GLUHFWLRQ DQG SRVLWLRQ RI WKH SL[HO ZLWK D FRQQHFWLRQ RI QHLJKERULQJ SL[HOV UHVXOWV LQ DQJOH 7KHVH WZR SURSHUWLHV PDGH *UDSK 6LJQDO 3URFHVVLQJ PRUH DGDSWDEOH IRU LPDJH GDWD ,Q WKH VLPXODWLRQ WKH SHUIRUPDQFH RI SURSRVHG PRGHO LV FRPSDUHG ZLWK WKH WKUHH FRQYHQWLRQDO ILOWHUV 9ROWHUUD ILOWHU PHGLDQ ILOWHU DQG VWDFN ILOWHUV 7KH SHUIRUPDQFH PHDVXUHPHQWV RI WKH PRGHOV DUH EDVHG RQ WKH 3HDN 6LJQDO WR 1RLVH 5DWLR 0HDQ 6TXDUH (UURU 5RRW PHDQV VTXDUH HUURU DQG 6WUXFWXUDO VLPLODULW\ LQGH[ .H\ZRUGVｲ *UDSK EDVHG ORZ SDVV ILOWHU 6DOW  SHSSHU QRLVH *DXVVLDQ QRLVH 9ROWHUUD )LOWHU 0HGLDQ )LOWHU DQG 6WDFN )LOWHU , ,1752'8&7,21 ,Q WKH LPDJHV WKH SUHVHQFH RI QRLVH FDQ EH GXH WR WKH LPDJH FDSWXULQJ VHQVRU RU GXH WR WKH SKRWRQ GHWHFWRU 7KHUH DUH GLIIHUHQW W\SHV RI QRLVH IRU H[DPSOH  ,PSXOVH QRLVH  *DXVVLDQ 1RLVH  6DOW DQG SHSSHU QRLVH >@ 7KH FODVVLFDO OLQHDU ILOWHUV DUH ZHOO VXLWHG IRU WKH UHGXFWLRQ DQG UHPRYDO RI WKH LPSXOVH QRLVHV RQO\ KRZHYHU WKHVH ILOWHUV FDQ EOXU WKH LPDJH GXULQJ WKH ILOWUDWLRQ SURFHVV DQG WKHVH ILOWHUV DUH LQHIIHFWLYH WRZDUGV WKH *DXVVLDQ QRLVHV 7KH\ ZHUH VWUDLJKWIRUZDUG WR FRQVWUXFW DQG LPSOHPHQW EHFDXVH RI WKHLU PDWKHPDWLFDO VLPSOLFLW\ DQG WKH SUHVHQFH RI VHYHUDO GHVLUDEOH FKDUDFWHULVWLFV ,Q DGGLWLRQ LQ D YDVW QXPEHU RI VFHQDULRV OLQHDU ILOWHUV IXQFWLRQHG EULOOLDQWO\ ,Q WKH HYHQW RI QRQ DGGLWLYH QRLVH DQG RI QRQOLQHDULW\ RI WKH V\VWHP RU *DXVVLDQ VWDWLVWLFV VFHQDULRV WKH QRLVH UHVXOWV DUH EDG >@ /LQHDU ILOWHUV DUH SRRU LQ UHGXFLQJ *DXVVLDQ DQG SRWDWR QRLVH LQ LPDJLQJ DSSOLFDWLRQV EHFDXVH WKH\ EOXU WKH HGJHV RI WKH ILOWHUV 7KH 9ROWHUUD VHULHV ZDV UHFRJQL]HG DV D PDMRU WRRO WR JHQHUDOL]H QRQOLQHDU V\VWHP WKHRU\ WKDW KDV OHG WR WKH SRO\QRPLDO ILOWHU RI QRQOLQHDU ILOWHUV >@>@ )RU H[DPSOH LQ >@ WKH 9ROWHUUD VHULHV LV WUXQFDWHG IRU WKH VHFRQG RUGHU WKLUG DQG ILIWK RUGHU XVHG IRU QRLVH UHGXFWLRQ IURP WHOHYLVLRQ LPDJHV DQG VKRZHG WKDW LQFUHDVLQJ RUGHU RI 9ROWHUUD VHULHV JLYHV D GLPLQLVKLQJ UHWXUQ $QG LQ >@ QG RUGHU 9ROWHUUD ILOWHU ZDV GHPRQVWUDWHG RQ DXGLR VLJQDO FRQWDLQLQJ *DXVVLDQ QRLVH 2QH RI WKH PDLQ SUREOHPV ZLWK WKH 9ROWHUUD ILOWHU LV LWV GHVLJQ EHFDXVH WKH 9ROWHUUD ILOWHUV KDYH VHYHUDO SDUDPHWHUV WKDW DUH UHTXLUHG WR HVWLPDWH WKH JURZWK H[SRQHQWLDOO\ ZLWK WKH RUGHU RI WKH H[SDQVLRQ 7KH PHGLDQ ILOWHUV DUH WKH W\SH RI OLQHDU ILOWHU DQG DUH PDLQO\ XVHG IRU WKH LPSXOVH W\SH RI QRLVH 7KLV ILOWHU FDQQRW EH XVHG IRU WKH *DXVVLDQ QRLVH EXW LW LV ZHOO VXLWHG IRU WKH VDOW DQG SHSSHU W\SHV RI QRLVH UHPRYDO 1HYHUWKHOHVV RQH RI WKH PRVW ZLGHO\ XVHG DSSURDFKHV LV PHGLDQ ILOWHULQJ >@ >@ >@ DQG >@ 7R FRQVWUXFW D  [  VHSDUDEOH PHGLDQ ILOWHU D ZLQGRZ ZLGWK  PHGLDQ ILOWHUV DUH DSSOLHG WR HDFK RI WKH LPDJHV URZV WR PDNH DQ LQWHUPHGLDWH LPDJH DQG WKHQ WKH PHGLDQ LV DSSOLHG WR HDFK RI WKH LQWHUPHGLDWH LPDJHV FROXPQV WR FUHDWH WKH ILQDO LPDJH %HFDXVH WKH FXEHV HGJH LV VKDUS \HW KDV D MLWWHU HGJH WKH PHGLDQ LV QRW VXLWDEOH IRU ILOWHULQJ UHVXOWLQJ LQ D QRQVTXDUH LPDJH 7KH HGJH MLWWHUV RI PHGLDQ ILOWHULQJ ZHUH LQYHVWLJDWHG LQ >@ DQG >@ :KHQ SLFWXUHV DUH ILOWHUHG E\ D PHGLDQ ILOWHU WKH SK\VLFDO HIIHFW RI VWUHDNLQJ RFFXUV >@ :H HPSOR\HG DQ LPDJH GHQRLVLQJ XVLQJ VWDFN DQG 9ROWHUUD ILOWHUV LQ WKLV VWXG\ 7KH ZHDN VXSHUSRVLWLRQ SURSHUW\ DOVR NQRZQ DV WKH GHFRPSRVLWLRQ WKUHVKROG >@ >@ DQG WKH SLFNLQJ SURSHUW\ >@ DOVR NQRZQ DV WKH VWDFNLQJ SURSHUW\ LQ >@ >@ DQG >@ DUH ERWK VKDUHG E\ VWDFN ILOWHUV %HIRUH WKH VWDQGDUG JUDGLHQW GHVFHQW HVWLPDWRU LV DSSOLHG WR WKH LPDJH IRU VPRRWKLQJ DQG HGJH GHWHFWLRQ VWDFN ILOWHUV DUH DSSOLHG WR WKH LPDJH WKLV LV D SUHILOWHULQJ W\SH ILOWHU DQG LW DOORZV WKH LPDJH UHVROXWLRQ LQWR ILQHU GHWDLO %XW VRPHWLPHV WKLV ILOWHU LV YHU\ VHQVLWLYH WR LPSXOVH QRLVH %\ DSSO\LQJ WKLV ILOWHU LW LV HDV\ WR REWDLQ WKH ORFDO HVWLPDWLRQ RI WKH SXUH DQG HQFRGH YHUVLRQ RI WKH FRUUXSWLRQ QRLVHIUHH LPDJH 7KLV ILOWHU ZRUNV ZHOO RQ WKH LPDJH ZLWK WKH DGGLWLYH *DXVVLDQ QRLVH WKDQ WKH LPDJH ZLWK WKH H[WUHPH LPSXOVH QRLVH $FFRUGLQJ WR WKH VWXG\ RI OLWHUDWXUH >@ WKH VWDFN ILOWHU LV YHU\ UREXVW DQG WKLV FKDUDFWHULVWLF ZLOO OHDG WR WKH RXWVWDQGLQJ 3HUIRUPDQFH RI WKH HOLPLQDWLRQ RI WKH WDUJHW QRLVH DQG UHGXFH WKH VWDWLVWLFDO QRLVH IURP GDWD DQG LPDJHV IURP WKH WUDLQLQJ 8VLQJ OH[LFRJUDSKLF EORFN PDWULFHV DQG V\PPHWU\ FULWHULD 9ROWHUUD LPDJH UHVWRUDWLRQ ILOWHUV DUH VXJJHVWHG >@ ,DQ - 0RUULVRQ DQG 3HWHU M: 5D\QHU >@ XVHG QRQOLQHDU :LHQHU ILOWHUV WR HYDOXDWH VLJQDOV FRQWDPLQDWHG E\ DGGLWLYH QRLVH E\ EURDGHQLQJ WKH OLQHDU ILOWHU WR QRQOLQHDU WHUPV 7KH GHYHORSPHQW RI WKHRU\ DQG PRGHOV WR DQDO\VH GDWD RQ LUUHJXODU GRPDLQV VXFK DV JUDSKV LV EHFRPLQJ PRUH SRSXODU LQ VLJQDO SURFHVVLQJ DQG PDFKLQH OHDUQLQJ 7\SLFDO JUDSKV LOOXVWUDWH UHODWLRQVKLSV IRU H[DPSOH VRFLDO HFRQRPLF DQG ELRORJLFDO QHWZRUNV ZKHUH VHYHUDO JUDSKEDVHG PDWK DQG VWDWLVWLF PRGHOV ZHUH GHYHORSHG WR H[SODLQ WKH WUHQGV LQ GDWD >@>@ $QRWKHU H[DPSOH LV SK\VLFDO LQIUDVWUXFWXUH VXFK DV SRZHU JDV DQG ZDWHU V\VWHPV QHWZRUNV DV ZHOO DV WUDQVSRUWDWLRQ QHWZRUNV ZKLFK KDV DQ LPSDFW RQ WKH VLJQDO VKDSH LQ DGGLWLRQ WR FRQQHFWLYLW\ SK\VLFDO SULQFLSOHV 6WDWLVWLFV UHOLHG KHDYLO\ XSRQ JUDSKLF LQWHUSUHWDWLRQ IRU D ORQJ WLPH WR FRQFOXGH JUDSKLF PRGHOV 7KH JUDSK GDWD SURFHVVLQJ VHFWRU >@ KDV EHHQ FUHDWHG E\ WKH QHHG WR DGGUHVV WKLV ZKLFK VWUHDPOLQHV WKH JUDSK GDWD FKDUDFWHULVWLFV DQG GULYHV VLJQDO SURFHVVLQJ WHFKQLTXHV ZLWK D GHWHUPLQLVWLF DQG V\VWHP WKHRUHWLFDO DSSURDFK $ FRUH RI *63 LV D IRUPDO JUDSK ILOWHU GHVFULSWLRQ ZKLFK DSSOLHV WKH SULQFLSOHV RI /7, ILOWHULQJ RI 2022 International Conference on Recent Trends in Microelectronics, Automation, Computing and Communications Systems (ICMACC) 2022 Interational Conference on Recent Trends in Microelectronics, Automation, Computing and Communications Systems (ICMACC) Image Filtration using Graph-based Low Pass Filter Nagaraj u Sonti Deparent ofECE Vignan 'sF oundation for Science, Technolog & Research Vadlamudi, Andhra Pradesh, India nagarajusonti@gmail.com Rukmini M S S Depart ent of ECE Vignan'sF oundation for Science, Technolog & Research Vadlamudi, Andhra Pradesh, India mssrukmini@gmail.com Venkatappa Reddy Pamulapati Depart ent of ECE Vignan 'sF oundation for Science, Technolog & Research Vadlamudi, Andhra Pradesh, India venkatappareddyp@gmail. com Abstract- This paper is based on fltering various noises from the image data through the graph-based low pass flter. The linear noise (salt & pepper noise) and nonlinear noise (Gaussian noise) are taken for the research purpose, those noises are randomly added to the image data. I this paper, the graph flter is proposed to remove the linear as well as non linear noise. This graph-based low pass flter also allows the noise from the bulk image dataset. I the graph-based low pass flter, the image dataset is converted into a structure like a graph and then the image is considered as matrix data. The pixels in the data connected with neighboring pixels that demonstrate the direction and position of the pixel with a connection of neighboring pixels results in angle. These two properties made Graph Signal Processing more adaptable for image data. I the simulation, the performance of proposed model is compared with the three conventional flters (olterra flter, median flter, and stack flters). The performance measurements of the models are based on the Peak Signal to Noise Ratio, Mean Square Error, Root means square error, and Structural similarity index. Keyord- Graph based lw pass filer, Sal & pepper noise, Gaussian noise, Volerra Filer, Median Filer and Stack Filer. I. INTRODUCTION In the images, the presence of noise can be due to the image capturing sensor or due to the photon detector. There are diferent tpes of noise for example 1) Impulse noise 2) Gaussian Noise 3) Salt and pepper noise [1]. The classical linear flters are well suited for the reduction and removal of the impulse noises only, however, these filters can blur the image durng the filtration process, and these filters are ineffective towards the Gaussian noises. They were straightforward to construct and implement because of their mathematical simplicit and the presence of several desirable characteristics. In addition, in a vast number of scenarios, linear flters fnctioned brilliantly. In the event of non additive noise and of nonlinearit of the system or Gaussian statistics scenarios, the noise results are bad [2]. Linear filters are poor in reducing Gaussian and potato noise in imaging applications because they blur the edges of the flters. The Volterra series was recognized as a major tool to generalize non-linear system theory that has led to the polynomial flter of nonlinear filters [3], [4]. For example, in [5], the Volterra series is truncated for the second order, third and fifth order used for noise reduction from television images and showed that increasing order of Volterra series gives a diminishing retur. And in [6], 2nd order Volterra filter was demonstrated on audio signal containing Gaussian noise. One of the main problems with the Volterra filter is its design because the Volterra filters have several parameters that are required to estimate the growth exponentially with the order of the exansion. The median flters are the tpe of linear flter and are mainly used for the impulse tpe of noise. This flter 978- 1-6654-9604-9/22/$31.00 ©2022 IEEE 125 cannot be used for the Gaussian noise, but it is well suited for the salt and pepper tpes of noise removal. Nevertheless, one of the most widely used approaches is median filtering [7], [8], [9], and [10]. To construct a 3 x 3 separable median filter, a window width 3 median filters are applied to each of the image's rows to make an intermediate image, and then the median is applied to each of the intermediate image's colunms to create the final image. Because the cube's edge is shar yet has a jitter edge, the median is not suitable for fltering, resulting in a non-square image. The edge jitters of median filtering were investigated in [11] and [12]. When pictures are filtered by a median flter, the physical effect of streaking occurs [13]. We employed an image denoising using stack and Volterra flters in this study. The weak superosition propert, also known as the decomposition threshold [14], [16], and the picking propert [15], also known as the stacking propert in [14], [16], and [17], are both shared by stack flters. Before the standard gradient descent estimator is applied to the image for smoothing and edge detection, stack filters are applied to the image this is a preflterng type filter and it allows the image resolution into finer detail. But sometimes this flter is very sensitive to impulse noise. By applying this flter, it is easy to obtain the local estimation of the pure and encode version of the corruption noise-free image. This flter works well on the image with the additive Gaussian noise than the image with the extreme impulse noise. According to the study of literature [1], the stack filter is very robust, and this characteristic will lead to the outstanding Perormance of the elimination of the target noise and reduce the statistical noise from data and images from the training. Using lexicographic block matrices and symmetry crteria, Volterra image restoration flters are suggested [18]. Ian J. Morrison and Peter j.W. Rayner [19] used nonlinear Wiener flters to evaluate signals contaminated by additive noise by broadening the linear flter to nonlinear terms. The development of theory and models to analyse data on irregular domains, such as graphs, is becoming more popular in signal processing and machine leang. Typical graphs illustrate relationships, for example social, economic, and biological networks, where several graph-based math and statistic models were developed to explain the trends in data [20], [21]. Another example is physical infrastructure (such as power, gas, and water systems networks as well as transportation networks) which has an impact on the signal shape, in addition to connectivit, physical principles. Statistics relied heavily upon graphic interretation for a long time to conclude graphic models. The graph data processing sector [22] has been created by the need to address this, which streamlines the graph data characteristics and drives signal processing techniques with a deterministic and system theoretical approach. A core of G SP is a formal graph flter description, which applies the principles of LTI fltering of Authorized licensed use limited to: Andhra University College of Engineering. Downloaded on June 20,2023 at 09:32:35 UTC from IEEE Xplore. Restrictions apply.