image processing - Algorithm to detect corners of paper sheet in photo -


what best way detect corners of invoice/receipt/sheet-of-paper in photo? used subsequent perspective correction, before ocr.

my current approach has been:

rgb > gray > canny edge detection thresholding > dilate(1) > remove small objects(6) > clear boarder objects > pick larges blog based on convex area. > [corner detection - not implemented]

i can't think there must more robust 'intelligent'/statistical approach handle type of segmentation. don't have lot of training examples, 100 images together.

broader context:

i'm using matlab prototype, , planning implement system in opencv , tesserect-ocr. first of number of image processing problems need solve specific application. i'm looking roll own solution , re-familiarize myself image processing algorithms.

here sample image i'd algorithm handle: if you'd take challenge large images @ http://madteckhead.com/tmp

case 1 http://madteckhead.com/tmp/img_0773_sml.jpg case 2 http://madteckhead.com/tmp/img_0774_sml.jpg case 3 http://madteckhead.com/tmp/img_0775_sml.jpg case 4 http://madteckhead.com/tmp/img_0776_sml.jpg

in best case gives:

case 1 - canny http://madteckhead.com/tmp/img_0773_canny.jpg case 1 - post canny http://madteckhead.com/tmp/img_0773_postcanny.jpg case 1 - largest blog http://madteckhead.com/tmp/img_0773_blob.jpg

however fails on other cases:

case 2 - canny http://madteckhead.com/tmp/img_0774_canny.jpg case 2 - post canny http://madteckhead.com/tmp/img_0774_postcanny.jpg case 2 - largest blog http://madteckhead.com/tmp/img_0774_blob.jpg

thanks in advance great ideas! love so!

edit: hough transform progress

q: algorithm cluster hough lines find corners? following advice answers able use hough transform, pick lines, , filter them. current approach rather crude. i've made assumption invoice less 15deg out of alignment image. end reasonable results lines if case (see below). not entirely sure of suitable algorithm cluster lines (or vote) extrapolate corners. hough lines not continuous. , in noisy images, there can parallel lines form or distance line origin metrics required. ideas?

case 1 http://madteckhead.com/tmp/img_0773_hough.jpg case 2 http://madteckhead.com/tmp/img_0774_hough.jpg case 3 http://madteckhead.com/tmp/img_0775_hough.jpg case 4 http://madteckhead.com/tmp/img_0776_hough.jpg

i'm martin's friend working on earlier year. first ever coding project, , kinda ended in bit of rush, code needs errr...decoding... i'll give few tips i've seen doing already, , sort code on day off tomorrow.

first tip, opencv , python awesome, move them possible. :d

instead of removing small objects , or noise, lower canny restraints, accepts more edges, , find largest closed contour (in opencv use findcontour() simple parameters, think used cv_retr_list). might still struggle when it's on white piece of paper, providing best results.

for houghline2() transform, try cv_hough_standard opposed cv_hough_probabilistic, it'll give rho , theta values, defining line in polar coordinates, , can group lines within tolerance those.

my grouping worked table, each line outputted hough transform give rho , theta pair. if these values within, 5% of pair of values in table, discarded, if outside 5%, new entry added table.

you can analysis of parallel lines or distance between lines more easily.

hope helps.


Comments

Post a Comment

Popular posts from this blog

c# - SharpSVN - How to get the previous revision? -

i trying find efficient way previous revision of file text comparison using sharpsvn. using (svnclient c = new svnclient()) { c.authentication.defaultcredentials = new networkcredential( configurationmanager.appsettings.get("svnserviceusername") , configurationmanager.appsettings.get("svnservicepassword") , configurationmanager.appsettings.get("svnservicedomain") ); c.authentication.sslservertrusthandlers += new eventhandler<svnsslservertrusteventargs>(authentication_sslservertrusthandlers); collection<svnfileversioneventargs> fileversioncollection = new collection<svnfileversioneventargs>(); svnrevisionrange range = new svnrevisionrange(0, this.hooks.revision); svnfileversionsargs args = new svnfileversionsargs(); args.retrieveproperties = true; args.range = range; foreach (svnchangeitem item in log.changedpaths) { string path = this.repositorypath + it...
Read more

c++ - Is it possible to compile a VST on linux? -

for class project i'm attempting write vst plugin backed cuda. current cuda workflow on linux box, i'd prefer compile , link there. according wikipedia , should possible (i couldn't find steinberg documentation relevant linux) can't find makefile or instructions on how build if aren't using xcode or visual studio. i'm vst 3 sdk doesn't support linux. when try compile plugin under linux, error: ./base/source/fatomic.cpp:39:30: fatal error: libkern/osatomic.h: no such file or directory this issue caused following code in "vst3 sdk/base/source/fatomic.cpp" #if mac #include <libkern/osatomic.h> #if mac_os_x_version_min_required > mac_os_x_version_10_4 #define native_atomic_type (volatile int32_t*) #else #define native_atomic_type (int32_t*) #endif #elif windows #include <windows.h> #endif but hope compiling under linux work vst sdk 2.4. reading. jvstwrapper seems run on lin...
Read more

url - Querystring manipulation of email Address in PHP -

i using third party application (email2http) send me emailid, body , message. unable retrieve email id due name appended on , enclosed < , >. here querystring getting. test.php?from=ajay+reddy &subject=cx&body=somebody when print variable in php returns name , not email id. $from = $_get['from']; echo $from; //output ajay reddy please suggest me how can emailid also. i dont want use post method due testing , debugging. it's working fine. forgot use htmlentities() in debugging.
Read more