From bcebb8f17360fab0f4d22178bef2d33851fef312 Mon Sep 17 00:00:00 2001
From: Samo Penic <samo.penic@gmail.com>
Date: Fri, 30 Nov 2018 07:58:51 +0000
Subject: [PATCH] Moved the answer ROIs few pixels. Recommended threshold around 0.3
---
aoi_ocr/Ocr.py | 44 +++++++++++++++++++++++++++-----------------
1 files changed, 27 insertions(+), 17 deletions(-)
diff --git a/aoi_ocr/Ocr.py b/aoi_ocr/Ocr.py
index f4b2eae..99ad414 100644
--- a/aoi_ocr/Ocr.py
+++ b/aoi_ocr/Ocr.py
@@ -5,7 +5,7 @@
import os
import pkg_resources
-markerfile = '/template.png' # always use slash
+markerfile = '/template-sq.png' # always use slash
markerfilename = pkg_resources.resource_filename(__name__, markerfile)
@@ -41,6 +41,7 @@
return
self.decodeQRandRotate()
self.imgTreshold()
+ cv2.imwrite('/tmp/debug_threshold.png', self.bwimg)
skewAngle = 0
# try:
# skewAngle=self.getSkewAngle()
@@ -96,10 +97,10 @@
self.imgHeight, self.imgWidth = self.img.shape[0:2]
# todo, make better tresholding
-
def imgTreshold(self):
(self.thresh, self.bwimg) = cv2.threshold(
- self.img, 128, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU
+ self.img, 128, 255,
+ cv2.THRESH_BINARY | cv2.THRESH_OTSU
)
def getSkewAngle(self):
@@ -135,7 +136,7 @@
def locateUpMarkers(self, threshold=0.85, height=200):
template = cv2.imread(markerfilename, 0)
w, h = template.shape[::-1]
- crop_img = self.img[0:height, :]
+ crop_img = self.bwimg[0:height, :]
res = cv2.matchTemplate(crop_img, template, cv2.TM_CCOEFF_NORMED)
loc = np.where(res >= threshold)
cimg = cv2.cvtColor(crop_img, cv2.COLOR_GRAY2BGR)
@@ -172,7 +173,8 @@
def locateRightMarkers(self, threshold=0.85, width=200):
template = cv2.imread(markerfilename, 0)
w, h = template.shape[::-1]
- crop_img = self.img[:, -width:]
+ crop_img = self.bwimg[:, -width:]
+ cv2.imwrite('/tmp/debug_right.png', crop_img)
res = cv2.matchTemplate(crop_img, template, cv2.TM_CCOEFF_NORMED)
loc = np.where(res >= threshold)
cimg = cv2.cvtColor(crop_img, cv2.COLOR_GRAY2BGR)
@@ -188,9 +190,13 @@
loc_filtered_y.append(pt[1])
loc_filtered_x.append(pt[0])
# order by y coordinate
- loc_filtered_y, loc_filtered_x = zip(
- *sorted(zip(loc_filtered_y, loc_filtered_x))
- )
+ try:
+ loc_filtered_y, loc_filtered_x = zip(
+ *sorted(zip(loc_filtered_y, loc_filtered_x))
+ )
+ except:
+ self.yMarkerLocations=[np.array([1,1]),np.array([1,2])]
+ return self.yMarkerLocations
# loc=[loc_filtered_y,loc_filtered_x]
# remove duplicates
a = np.diff(loc_filtered_y) > 40
@@ -210,12 +216,12 @@
self.locateUpMarkers()
self.locateRightMarkers()
- roixoff = 10
- roiyoff = 5
- roiwidth = 50
+ roixoff = 4
+ roiyoff = 0
+ roiwidth = 55
roiheight = roiwidth
totpx = roiwidth * roiheight
-
+ cimg = cv2.cvtColor(self.img, cv2.COLOR_GRAY2BGR)
self.answerMatrix = []
for y in self.yMarkerLocations[0]:
oneline = []
@@ -227,6 +233,8 @@
# cv2.imwrite('ans_x'+str(x)+'_y_'+str(y)+'.png',roi)
black = totpx - cv2.countNonZero(roi)
oneline.append(black / totpx)
+ cv2.rectangle(cimg, (x - roixoff,y - roiyoff), (x + int(roiwidth - roixoff),y + int(roiheight - roiyoff)), (0, 255, 255), 2)
+ cv2.imwrite('/tmp/debug_answers.png',cimg)
self.answerMatrix.append(oneline)
def get_enhanced_sid(self):
@@ -237,7 +245,7 @@
es, err, warn = getSID(
self.img[
int(0.04 * self.imgHeight) : int(0.095 * self.imgHeight),
- int(0.7 * self.imgWidth) : int(0.99 * self.imgWidth),
+ int(0.65 * self.imgWidth) : int(0.95 * self.imgWidth),
],
self.sid_classifier,
sid_mask,
@@ -270,9 +278,10 @@
data = qrdata.split(",")
retval = {
"exam_id": int(data[1]),
- "page_no": int(data[3])+1,
+ "page_no": int(data[3]),
"paper_id": int(data[2]),
"faculty_id": int(data[0]),
+ "sid": None
}
if len(data) > 4:
retval["sid"] = data[4]
@@ -285,11 +294,11 @@
data["errors"] = self.errors
data["warnings"] = self.warnings
data["up_position"] = (
- list(self.xMarkerLocations[1] / self.imgWidth),
- list(self.yMarkerLocations[1] / self.imgHeight),
+ list(self.xMarkerLocations[0] / self.imgWidth),
+ list(self.xMarkerLocations[1] / self.imgHeight),
)
data["right_position"] = (
- list(self.xMarkerLocations[1] / self.imgWidth),
+ list(self.yMarkerLocations[0] / self.imgWidth),
list(self.yMarkerLocations[1] / self.imgHeight),
)
data["ans_matrix"] = (
@@ -300,4 +309,5 @@
output_filename=os.path.join(self.output_path, '.'.join(self.filename.split('/')[-1].split('.')[:-1])+".png")
cv2.imwrite(output_filename, self.img)
data['output_filename']=output_filename
+ print(np.array(self.answerMatrix))
return data
--
Gitblit v1.9.3