|
Hi I am just learning image processing. I want to implement edge detection by sobel operator in 'c', i am facing some problems. i am first reading image. then i am applying first mask to it and gradient(let gx) in another veriable. now i am applying second mask to original image and saving that gradient(let gy). now my final value of gradient is g=|gx|+|gy| (as per approximate eq) my que are 1. what should we do if value of g is greater than max gray level (in my case it is 255)? 2. value of gx or gy may be -ve so is it required to do some scaling to it before adding? 3. is we have to divide by some factor (like 9 in 3x3 low pass filter where all mask elements have value 1) while applying mask? 4. does sobel or prewitt operator have some limitations regarding type of image? can i use these operators to find edge of human face (eg std. image of lena) i am facing some prob. i have implemented sobel operator. i download some images from net and use them to find edge detection but the resultant image have blurring effect where as on net i found highly enhanced image for same image so i am not getting where i am wrong........ so plz help me if u can....... thank in advance Snehal |
|
|
about edge detection
Started by ●September 21, 2002
Reply by ●September 23, 20022002-09-23
|
gradient magnitude is given by |G| = (Gx^2 + Gy^2)^1/2 Although typically, an approximate magnitude is computed using: |G| = |Gx| + |Gy| It is faster to compute. Output values from the operator can easily overflow the maximum allowed pixel value for image types that only support smallish integer pixel values (e.g. 8-bit integer images). When this happens the standard practice is to simply set overflowing output pixels to the maximum allowed value. The problem can be avoided by using an image type that supports pixel values with a larger range. i.e You can make magnitudes >255 = 255. Good Luck, Sayaji >From: "Snehal" <> >To: >Subject: [imagedsp] about edge detection >Date: Sat, 21 Sep 2002 11:32:07 -0000 >MIME-Version: 1.0 >X-Originating-IP: 203.200.42.210 >Received: from n15.grp.scd.yahoo.com ([66.218.66.70]) by >mc4-f34.law16.hotmail.com with Microsoft SMTPSVC(5.0.2195.5600); Mon, 23 >Sep 2002 06:41:40 -0700 >Received: from [66.218.67.192] by n15.grp.scd.yahoo.com with NNFMP; 23 Sep >2002 13:42:01 -0000 >Received: (qmail 42832 invoked from network); 23 Sep 2002 13:41:58 -0000 >Received: from unknown (66.218.66.216) by m10.grp.scd.yahoo.com with QMQP; >23 Sep 2002 13:41:58 -0000 >Received: from unknown (HELO n28.grp.scd.yahoo.com) (66.218.66.84) by >mta1.grp.scd.yahoo.com with SMTP; 23 Sep 2002 13:41:58 -0000 >Received: from [66.218.67.154] by n28.grp.scd.yahoo.com with NNFMP; 23 Sep >2002 13:41:58 -0000 >Received: (EGP: mail-8_1_1_3); 21 Sep 2002 11:32:08 -0000 >Received: (qmail 2898 invoked from network); 21 Sep 2002 11:32:08 -0000 >Received: from unknown (66.218.66.218) by m7.grp.scd.yahoo.com with QMQP; >21 Sep 2002 11:32:08 -0000 >Received: from unknown (HELO n9.grp.scd.yahoo.com) (66.218.66.93) by >mta3.grp.scd.yahoo.com with SMTP; 21 Sep 2002 11:32:08 -0000 >Received: from [66.218.67.174] by n9.grp.scd.yahoo.com with NNFMP; 21 Sep >2002 11:32:08 -0000 >X-eGroups-Return: > >X-eGroups-Return: >X-Sender: >X-Apparently-To: >Message-ID: <> >User-Agent: eGroups-EW/0.82 >X-Mailer: Yahoo Groups Message Poster >X-Yahoo-Profile: snehal_joshi007 >X-eGroups-Approved-By: s_boucheryah <> via web; 23 >Sep 2002 13:41:56 -0000 >Mailing-List: list ; contact > >Delivered-To: mailing list >Precedence: bulk >List-Unsubscribe: <mailto:> >Return-Path: > >X-OriginalArrivalTime: 23 Sep 2002 13:41:40.0125 (UTC) >FILETIME=[F20654D0:01C26306] > >Hi >I am just learning image processing. >I want to implement edge detection by sobel operator in 'c', i am >facing some problems. i am first reading image. then i am applying >first mask to it and gradient(let gx) in another veriable. now i am >applying second mask to original image and saving that gradient(let >gy). >now my final value of gradient is g=|gx|+|gy| (as per approximate eq) > >my que are >1. what should we do if value of g is greater than max gray level (in >my case it is 255)? >2. value of gx or gy may be -ve so is it required to do some scaling >to it before adding? >3. is we have to divide by some factor (like 9 in 3x3 low pass filter >where all mask elements have value 1) while applying mask? >4. does sobel or prewitt operator have some limitations regarding >type of image? can i use these operators to find edge of human face >(eg std. image of lena) > >i am facing some prob. i have implemented sobel operator. i download >some images from net and use them to find edge detection but the >resultant image have blurring effect where as on net i found highly >enhanced image for same image > >so i am not getting where i am wrong........ > >so plz help me if u can....... >thank in advance > >Snehal > _________________________________________________________________ MSN Photos is the easiest way to share and print your photos: http://photos.msn.com/support/worldwide.aspx |
Reply by ●September 24, 20022002-09-24
|
Hi, > I am just learning image processing. > I want to implement edge detection by sobel operator in 'c', i am > facing some problems. i am first reading image. then i am applying > first mask to it and gradient(let gx) in another veriable. now i am > applying second mask to original image and saving that gradient(let > gy). > now my final value of gradient is g=|gx|+|gy| (as per approximate eq) > > my que are > 1. what should we do if value of g is greater than max gray level (in > my case it is 255)? Well to the best of my knolwedge, when you are applying an operator on an image whose values range between 0 and 255, the result will never exceed 255 or the maximum value that an image can take. In case it doing so either you are not balancing the coefficients properly (which I think is the case) else u need to threshold the result ( which I dont think can happen) > 2. value of gx or gy may be -ve so is it required to do some scaling > to it before adding? Same as above. > 3. is we have to divide by some factor (like 9 in 3x3 low pass filter > where all mask elements have value 1) while applying mask? see when u apply a median filter in spatial domain, the sum of the coefficients will be 1, whereas for Sobel and Prewitt operators, the sum of coefficients in the mask will be 0 (Ref. Pg. 349 Sec.9.4. Fundamental of Digital Image Processing by A K Jain) > 4. does sobel or prewitt operator have some limitations regarding > type of image? can i use these operators to find edge of human face > (eg std. image of lena) See the edge detection operators have the property that they yield zeroes for uniform regions. Also I am not too sure how you are finding the edge map... Are u thresholding after you add the gradients.basically you should get a binary image where a pixel can either be 0 or 255. > i am facing some prob. i have implemented sobel operator. i download > some images from net and use them to find edge detection but the > resultant image have blurring effect where as on net i found highly > enhanced image for same image > > so i am not getting where i am wrong........ Hope my answers help you but I suggest you try some morphological gradients for good edge detection. Do revert in case you require anymore help. Ganesh *********************************** Ganesh Vijayan DSP Desgin Engineer - Video Emuzed India (P) Ltd. Bangalore www.emuzed.com ************************************* |
Reply by ●September 25, 20022002-09-25
|
>Hi, >> I am just learning image processing. >> I want to implement edge detection by sobel operator in 'c', i am >> facing some problems. i am first reading image. then i am applying >> first mask to it and gradient(let gx) in another veriable. now i am >> applying second mask to original image and saving that gradient(let >> gy). >> now my final value of gradient is g=|gx|+|gy| (as per approximate eq) >> >> my que are >> 1. what should we do if value of g is greater than max gray level (in >> my case it is 255)? > >Well to the best of my knolwedge, when you are applying an operator on an >image whose values range between 0 and 255, the result will never exceed 255 >or the maximum value that an image can take. In case it doing so either you >are not balancing the coefficients properly (which I think is the case) else >u need to threshold the result ( which I dont think can happen) > How about convolving [0 0 0 0 255 0 0 0 0] with [-1 2 -1] ? > >> 2. value of gx or gy may be -ve so is it required to do some scaling >> to it before adding? > >Same as above. > -ve signifies an 'opposite' edge (ie dark to lighht vs light to dark. <snip> >> i am facing some prob. i have implemented sobel operator. i download >> some images from net and use them to find edge detection but the >> resultant image have blurring effect where as on net i found highly >> enhanced image for same image >> >> so i am not getting where i am wrong........ Post some of your outputs so we can have a look. Also, if you have Paintshop PRo you can do a Sobel (eg) in that for comparison purposes. Cheers, Martin -- Martin Thompson BEng(Hons) CEng MIEE TRW Conekt Stratford Road, Solihull, B90 4GW. UK Tel: +44 (0)121-627-3569 - |
Reply by ●September 25, 20022002-09-25
|
Martin- > >Well to the best of my knolwedge, when you are applying an operator on an > >image whose values range between 0 and 255, the result will never exceed 255 > >or the maximum value that an image can take. In case it doing so either you > >are not balancing the coefficients properly (which I think is the case) else > >u need to threshold the result ( which I dont think can happen) > > > > How about convolving [0 0 0 0 255 0 0 0 0] with [-1 2 -1] ? Thanks for the morning laugh. Not to mention that practical DSP/fixed-point implementations are filled with fun issues like overflow, saturation, guard bands, etc. Jeff Brower DSP sw/hw engineer Signalogic |
