Merge pull request #2 from AeroAndZero/AdvanceOmrScanner

AOS v1

README.md

@@ -1,4 +1,2 @@
-# Simple OMR Scanner
- An Simple OMR Scanner Made Using OpenCV and Python.
-
-[Click Here](https://youtu.be/rMkkUP3v9h0) For A Youtube Video On Explaining How This Works. (Its an Unlisted Video.. You Can't See It Without Clicking The Link.)
+# Development Branch
+### This branch is only for in-progress code and not for download ! Switch to Default/master branch for stable application code.

aos.py

@@ -0,0 +1,201 @@
+import cv2
+import numpy as np
+import circleProcessor
+import random
+red = [10,10,255]
+blue = [255,10,10]
+green = [10,255,10]
+yellow = [0,255,255]
+purple = [255,10,255]
+answerKeyTextFile = 0
+
+def mapp(h):
+        h = h.reshape((4,2))
+        hnew = np.zeros((4,2),dtype = np.float32)
+
+        add = h.sum(1)
+        hnew[0] = h[np.argmin(add)]
+        hnew[2] = h[np.argmax(add)]
+
+        diff = np.diff(h,axis = 1)
+        hnew[1] = h[np.argmin(diff)]
+        hnew[3] = h[np.argmax(diff)]
+
+        return hnew
+
+def findCorners(omr0,cp1=70,cp2=20,bp=17):
+    h = omr0.shape[0]
+    w = omr0.shape[1]
+    xdownscale = float(500)/float(w)
+    ydownscale = float(500)/float(h)
+
+    copy = np.copy(omr0)
+    omr0 = cv2.resize(omr0,(int(w*xdownscale),int(h*ydownscale)))
+
+    omr0_blur = cv2.GaussianBlur(omr0,(bp,bp),0)
+
+    omr0_canny = cv2.Canny(omr0_blur,cp1,cp2)
+
+    contours, hierarchy = cv2.findContours(omr0_canny,cv2.RETR_TREE,cv2.CHAIN_APPROX_NONE)
+
+    cv2.drawContours(omr0, contours, -1, (0,255,0), 3)
+    maxArea = 0
+    maxContour = contours[0]
+    perimulti = 0.01
+    for i in range(0,len(contours)):
+        area = cv2.contourArea(contours[i])
+        if area == (omr0.shape[0]-1)*(omr0.shape[1]-1):
+            continue
+
+        temp_peri = cv2.arcLength(contours[i],True)
+        temp_approx = cv2.approxPolyDP(contours[i],perimulti*temp_peri,True)
+
+        if area > maxArea and len(temp_approx) == 4:
+            maxArea = area
+            maxContour = contours[i]
+            maxContourIndex = i
+
+    peri = cv2.arcLength(maxContour,True)
+    approx = cv2.approxPolyDP(maxContour,perimulti*peri,True)
+
+    corners = mapp(approx)
+    for corner in corners:
+        corner[0] = int(w*corner[0]/500)
+        corner[1] = int(h*corner[1]/500)
+        #cv2.circle(copy,(corner[0],corner[1]),2,(0,0,255),7)
+
+
+    wscale = abs(max(corners[0][0] - corners[1][0],corners[2][0] - corners[3][0]))
+    hscale = abs(max(corners[0][1] - corners[2][1],corners[1][1] - corners[3][1]))
+
+    dst = np.array([
+            [0, 0],
+            [wscale - 1, 0],
+            [wscale - 1, hscale - 1],
+            [0, hscale - 1]], dtype = "float32")
+
+    M = cv2.getPerspectiveTransform(corners,dst)
+    wrapped = cv2.warpPerspective(copy, M, (wscale, hscale))
+    return wrapped
+
+def scanOmr(image,actualSize = [],init = [],diff = [],resize = [],totalMCQs = 10,totalOptions = 4,showDots = False, method=0,InkThreshold = 120):
+    #Debug
+    print(InkThreshold+10)
+    #some parameters
+    safeToScan = False
+    actualw = actualSize[0]
+    actualh = actualSize[1]
+    fx = init[0]
+    fy = init[1]
+    dx = diff[0]
+    dy = diff[1]
+    currentOption = 1
+
+    # 0: total scanned, 1: correct MCQs, 2 : wrong mcqs
+    optionTicked = []
+
+    if (method == 0):   #----------------- Simple Scanning Algorithm
+        resizew = resize[0]
+        resizeh = resize[1]
+
+        dx = int((dx*resizew)/actualw)
+        dy = int((dy*resizeh)/actualh)
+        fx = int((fx*resizew)/actualw)
+        fy = int((fy*resizeh)/actualh)
+
+        image= cv2.resize(image,(resizew,resizeh))
+
+        for y in range(fy,fy+(dy*totalMCQs),dy):
+
+            optionTicked.append(0)
+            currentOption = 1
+
+            for x in range(fx,fx+(dx*totalOptions),dx):
+
+                if (image[y,x][0] < InkThreshold) and (image[y,x][1] < InkThreshold) and (image[y,x][2] < InkThreshold):
+                    print("Low inkThreshold found")
+
+                    #Question Number : int(y/dy)
+                    #Option Number : int(x/dx)+1
+
+                    cv2.circle(image,(x,y),int(dx*0.1),yellow,int(dx*0.15))
+
+                    if optionTicked[len(optionTicked)-1] == 0:
+                        optionTicked[len(optionTicked)-1] += currentOption
+                    else :
+                        optionTicked[len(optionTicked)-1] += 69
+
+                else:
+                    cv2.circle(image,(x,y),int(dx*0.1),purple,int(dx*0.15))
+
+                currentOption += 1
+
+        image = cv2.resize(image,(actualSize[0],actualSize[1]))
+        return image, optionTicked
+
+    elif (method == 1): #----------------- Circle Detection and Dynamic Difference Algorithm
+        image = cv2.resize(image,(actualSize[0],actualSize[1]))
+        fx,fy = circleProcessor.findClosestCircle(image,fx,fy,int(dx/2))
+        PointHistoryX = [fx]
+        PointHistoryY = [fy]
+
+        try:
+            for mcq in range(totalMCQs):
+                y = circleProcessor.findClosestCircle(image,PointHistoryX[0],PointHistoryY[0]+dy,int(dy/2))[1]
+
+                PointHistoryX = [fx]
+
+                optionTicked.append(0)
+
+                for option in range(totalOptions):
+                    x = circleProcessor.findClosestCircle(image,PointHistoryX[0]+dx,PointHistoryY[0],int(dx/2))[0]
+                    #Scanning Here
+                    xForScan = PointHistoryX[0]
+                    yForScan = PointHistoryY[0]
+
+                    #----------- Scanning
+                    if (image[yForScan,xForScan][0] < InkThreshold) and (image[yForScan,xForScan][1] < InkThreshold) and (image[yForScan,xForScan][2] < InkThreshold):
+                        print("Low inkThreshold found")
+
+                        #Question Number : int(y/dy)
+                        #Option Number : int(x/dx)+1
+
+                        cv2.circle(image,(xForScan,yForScan),1,yellow,2)
+
+                        if optionTicked[len(optionTicked)-1] == 0:
+                            optionTicked[len(optionTicked)-1] += option+1
+                        else :
+                            optionTicked[len(optionTicked)-1] += 69
+
+                    else:
+                        cv2.circle(image,(xForScan,yForScan),1,purple,2)
+
+                    #Keeping Data Short
+                    PointHistoryX.insert(0,x)
+                    dx = abs(PointHistoryX[0] - PointHistoryX[1])    #Dynamic Difference
+                    PointHistoryX.pop()
+
+                PointHistoryY.insert(0,y)
+                dy = abs(PointHistoryY[0] - PointHistoryY[1])
+                PointHistoryY.pop()
+
+        except Exception as e:
+            print(e)
+
+        return image, optionTicked
+
+    else:
+        print("Invalid Method Provided.")
+        return 0
+
+
+def main():
+    omr = cv2.imread("images/img_3.jpg")
+    found_omr = findCorners(omr)
+    answers = scanOmr(found_omr,[278,503],[27,24],[32,24],[278,503],20,4,True)
+    cv2.imshow("AOSv2 1",omr)
+    cv2.waitKey(0)
+    cv2.destroyAllWindows()
+
+if __name__ == "__main__":
+    main()

circleProcessor.py

@@ -0,0 +1,78 @@
+'''
+best circle detection parameters found yet :
+cv2.HoughCircles(smallImage_blur_gray, cv2.HOUGH_GRADIENT, 1.5, hcDistant,param1=100,param2=20)
+'''
+import cv2
+import numpy as np
+import math
+show = 0
+
+def cropImage(image,refX,refY):
+    limitX = int((image.shape[1] * 0.05))
+    limitY = int((image.shape[0] * 0.05))
+    rightX = 0
+    leftX = 0
+    upY = 0
+    downY = 0
+
+    while (rightX + refX < image.shape[1]) and (rightX < limitX):
+        rightX += 1
+
+    while (refX - leftX > 0) and (leftX < limitX):
+        leftX += 1
+
+    while (refY + downY < image.shape[0]) and (downY < limitY):
+        downY += 1
+
+    while (refY - upY > 0) and (upY < limitY):
+        upY += 1
+
+    croppedImage = image[(refY-upY):(refY+downY),(refX-leftX):(refX+rightX)]
+    return croppedImage
+
+def distants(x1,x2,y1,y2):
+    dis = math.sqrt((x2-x1)*(x2-x1) + (y2-y1)*(y2-y1))
+    return dis
+
+def findClosestCircle(image,refX,refY,hcDistant):
+    minInit = 0
+    smallImage = cropImage(image,refX,refY)
+    #smallImage_blur = cv2.GaussianBlur(smallImage,(1,1),1)
+    smallImage_blur_gray = cv2.cvtColor(smallImage,cv2.COLOR_BGR2GRAY)
+
+    newX = int(smallImage.shape[1]/2)
+    newY = int(smallImage.shape[0]/2)
+    diffX = refX - newX
+    diffY = refY - newY
+
+    circles = cv2.HoughCircles(smallImage_blur_gray, cv2.HOUGH_GRADIENT, 1.5, hcDistant,param1=100,param2=20)
+
+    if circles is not None:
+        circles = np.round(circles[0, :]).astype("int")
+
+        minDistantXY = [circles[0][0],circles[0][1]]
+        for (x, y, r) in circles:
+            if show:  #Debugging
+                cv2.circle(smallImage,(x,y),1,(0,0,255),2) 
+
+            if distants(x,newX,y,newY) <= distants(minDistantXY[0],newX,minDistantXY[1],newY):
+                minDistantXY = [x,y]
+
+        newX = minDistantXY[0]
+        newY = minDistantXY[1]
+
+    finalX = newX + diffX
+    finalY = newY + diffY
+
+    '''For Debugging '''
+    if show:
+        cv2.circle(smallImage,(newX,newY),2,(0,255,0),3)
+        cv2.imshow("Lo",smallImage)
+        cv2.waitKey(0)
+
+    return [finalX,finalY]
+
+if __name__ == "__main__":
+    image = cv2.imread("images/croppedOMR7-ticked.jpg")
+    show = 1
+    findClosestCircle(image,401,714,1)