Creating trackbars to scroll large image in OpenCV Python

Question

I am trying to create scrollbars in a window created by OpenCv python. I know that I need to implement the code to handle the scrolling/panning process but I have no idea where to start and I've looked everywhere. It is essential that I create the scrollbars in the OpenCV window instead of using some other GUI window framework. Below is the code I am using to load an image and scale the image(which works). Any help is appreciated. And please don't refer me to the opencv documentation on creating trackbars, I've read it and it doesn't help at all. Thanks!

import cv2
import cv2.cv as cv
import numpy as np

cv.NamedWindow('image', cv.CV_WINDOW_AUTOSIZE)
cv.NamedWindow('Control Window', cv.CV_WINDOW_AUTOSIZE)


print " Zoom In-Out demo "
print " Press u to zoom "
print " Press d to zoom "

img = cv2.imread('picture.jpg')


while(1):
    h,w = img.shape[:2]

    cv2.imshow('image',img)
    k = cv2.waitKey(10)

    if k==27 :
        break

    elif k == ord('u'):  # Zoom in, make image double size
        img = cv2.pyrUp(img,dstsize = (2*w,2*h))

    elif k == ord('d'):  # Zoom down, make image half the size
        img = cv2.pyrDown(img,dstsize = (w/2,h/2))

cv2.destroyAllWindows()

Answer 1

I had the same need, so today I created a class from scratch that handles mouse clicks, pan, and zoom on an OpenCV window. It works like this:

1. right-drag up or down to zoom
2. right-click to center the view on the mouse
3. drag the x and y trackbars to scroll
4. when you initialize it, you can optionally pass in a function that will be called when the user left-clicks on a pixel

(As far as I can tell, OpenCV can't read the mouse wheel and can't create a vertical trackbar, so the user experience is a little non-intuitive but it works.)

# -*- coding: utf-8 -*-
import cv2
import numpy as np

class PanZoomWindow(object):
    """ Controls an OpenCV window. Registers a mouse listener so that:
        1. right-dragging up/down zooms in/out
        2. right-clicking re-centers
        3. trackbars scroll vertically and horizontally 
    You can open multiple windows at once if you specify different window names.
    You can pass in an onLeftClickFunction, and when the user left-clicks, this 
    will call onLeftClickFunction(y,x), with y,x in original image coordinates."""
    def __init__(self, img, windowName = 'PanZoomWindow', onLeftClickFunction = None):
        self.WINDOW_NAME = windowName
        self.H_TRACKBAR_NAME = 'x'
        self.V_TRACKBAR_NAME = 'y'
        self.img = img
        self.onLeftClickFunction = onLeftClickFunction
        self.TRACKBAR_TICKS = 1000
        self.panAndZoomState = PanAndZoomState(img.shape, self)
        self.lButtonDownLoc = None
        self.mButtonDownLoc = None
        self.rButtonDownLoc = None
        cv2.namedWindow(self.WINDOW_NAME, cv2.WINDOW_NORMAL)
        self.redrawImage()
        cv2.setMouseCallback(self.WINDOW_NAME, self.onMouse)
        cv2.createTrackbar(self.H_TRACKBAR_NAME, self.WINDOW_NAME, 0, self.TRACKBAR_TICKS, self.onHTrackbarMove)
        cv2.createTrackbar(self.V_TRACKBAR_NAME, self.WINDOW_NAME, 0, self.TRACKBAR_TICKS, self.onVTrackbarMove)
    def onMouse(self,event, x,y,_ignore1,_ignore2):
        """ Responds to mouse events within the window. 
        The x and y are pixel coordinates in the image currently being displayed.
        If the user has zoomed in, the image being displayed is a sub-region, so you'll need to
        add self.panAndZoomState.ul to get the coordinates in the full image."""
        if event == cv2.EVENT_MOUSEMOVE:
            return
        elif event == cv2.EVENT_RBUTTONDOWN:
            #record where the user started to right-drag
            self.mButtonDownLoc = np.array([y,x])
        elif event == cv2.EVENT_RBUTTONUP and self.mButtonDownLoc is not None:
            #the user just finished right-dragging
            dy = y - self.mButtonDownLoc[0]
            pixelsPerDoubling = 0.2*self.panAndZoomState.shape[0] #lower = zoom more
            changeFactor = (1.0+abs(dy)/pixelsPerDoubling)
            changeFactor = min(max(1.0,changeFactor),5.0)
            if changeFactor < 1.05:
                dy = 0 #this was a click, not a draw. So don't zoom, just re-center.
            if dy > 0: #moved down, so zoom out.
                zoomInFactor = 1.0/changeFactor
            else:
                zoomInFactor = changeFactor
#            print("zoomFactor: %s"%zoomFactor)
            self.panAndZoomState.zoom(self.mButtonDownLoc[0], self.mButtonDownLoc[1], zoomInFactor)
        elif event == cv2.EVENT_LBUTTONDOWN:
            #the user pressed the left button. 
            coordsInDisplayedImage = np.array([y,x])
            if np.any(coordsInDisplayedImage < 0) or np.any(coordsInDisplayedImage > self.panAndZoomState.shape[:2]):
                print("you clicked outside the image area")
            else:
                print("you clicked on %s within the zoomed rectangle"%coordsInDisplayedImage)
                coordsInFullImage = self.panAndZoomState.ul + coordsInDisplayedImage
                print("this is %s in the actual image"%coordsInFullImage)
                print("this pixel holds %s, %s"%(self.img[coordsInFullImage[0],coordsInFullImage[1]]))
                if self.onLeftClickFunction is not None:
                    self.onLeftClickFunction(coordsInFullImage[0],coordsInFullImage[1])
        #you can handle other mouse click events here
    def onVTrackbarMove(self,tickPosition):
        self.panAndZoomState.setYFractionOffset(float(tickPosition)/self.TRACKBAR_TICKS)
    def onHTrackbarMove(self,tickPosition):
        self.panAndZoomState.setXFractionOffset(float(tickPosition)/self.TRACKBAR_TICKS)
    def redrawImage(self):
        pzs = self.panAndZoomState
        cv2.imshow(self.WINDOW_NAME, self.img[pzs.ul[0]:pzs.ul[0]+pzs.shape[0], pzs.ul[1]:pzs.ul[1]+pzs.shape[1]])

class PanAndZoomState(object):
    """ Tracks the currently-shown rectangle of the image.
    Does the math to adjust this rectangle to pan and zoom."""
    MIN_SHAPE = np.array([50,50])
    def __init__(self, imShape, parentWindow):
        self.ul = np.array([0,0]) #upper left of the zoomed rectangle (expressed as y,x)
        self.imShape = np.array(imShape[0:2])
        self.shape = self.imShape #current dimensions of rectangle
        self.parentWindow = parentWindow
    def zoom(self,relativeCy,relativeCx,zoomInFactor):
        self.shape = (self.shape.astype(np.float)/zoomInFactor).astype(np.int)
        #expands the view to a square shape if possible. (I don't know how to get the actual window aspect ratio)
        self.shape[:] = np.max(self.shape) 
        self.shape = np.maximum(PanAndZoomState.MIN_SHAPE,self.shape) #prevent zooming in too far
        c = self.ul+np.array([relativeCy,relativeCx])
        self.ul = (c-self.shape/2).astype(np.int)
        self._fixBoundsAndDraw()
    def _fixBoundsAndDraw(self):
        """ Ensures we didn't scroll/zoom outside the image. 
        Then draws the currently-shown rectangle of the image."""
#        print("in self.ul: %s shape: %s"%(self.ul,self.shape))
        self.ul = np.maximum(0,np.minimum(self.ul, self.imShape-self.shape))
        self.shape = np.minimum(np.maximum(PanAndZoomState.MIN_SHAPE,self.shape), self.imShape-self.ul)
#        print("out self.ul: %s shape: %s"%(self.ul,self.shape))
        yFraction = float(self.ul[0])/max(1,self.imShape[0]-self.shape[0])
        xFraction = float(self.ul[1])/max(1,self.imShape[1]-self.shape[1])
        cv2.setTrackbarPos(self.parentWindow.H_TRACKBAR_NAME, self.parentWindow.WINDOW_NAME,int(xFraction*self.parentWindow.TRACKBAR_TICKS))
        cv2.setTrackbarPos(self.parentWindow.V_TRACKBAR_NAME, self.parentWindow.WINDOW_NAME,int(yFraction*self.parentWindow.TRACKBAR_TICKS))
        self.parentWindow.redrawImage()
    def setYAbsoluteOffset(self,yPixel):
        self.ul[0] = min(max(0,yPixel), self.imShape[0]-self.shape[0])
        self._fixBoundsAndDraw()
    def setXAbsoluteOffset(self,xPixel):
        self.ul[1] = min(max(0,xPixel), self.imShape[1]-self.shape[1])
        self._fixBoundsAndDraw()
    def setYFractionOffset(self,fraction):
        """ pans so the upper-left zoomed rectange is "fraction" of the way down the image."""
        self.ul[0] = int(round((self.imShape[0]-self.shape[0])*fraction))
        self._fixBoundsAndDraw()
    def setXFractionOffset(self,fraction):
        """ pans so the upper-left zoomed rectange is "fraction" of the way right on the image."""
        self.ul[1] = int(round((self.imShape[1]-self.shape[1])*fraction))
        self._fixBoundsAndDraw()

if __name__ == "__main__":
    infile = "./testImage.png"
    myImage = cv2.imread(infile,cv2.IMREAD_ANYCOLOR)
    window = PanZoomWindow(myImage, "test window")
    key = -1
    while key != ord('q') and key != 27: # 27 = escape key
        #the OpenCV window won't display until you call cv2.waitKey()
        key = cv2.waitKey(5) #User can press 'q' or ESC to exit.
    cv2.destroyAllWindows()

Answer 2

Because I am doing image processing on the image like getting pixel information and I will loose that ability if I encapsulate it in a widget or window provided by the GUI framework

That isn't true. You could always update the image after doing your processing. For example look here and here especially.
These examples process images in OpenCv and put them in a PyQt gui frame. I am sure that you could do similar things with other Gui frameworks (I couldn't find anything for Tkinter). I think I have seen wxPython integrated in the past.

When you are making your program, be sure to display a copy of the image. that way, the image object will continue to be changeable, and you can just update the image in the Gui. For example, here is some pseudo-code:

image=Image("myimage.png")
image.resize(100,400)
img=QImage(image)#similar to how pyqt would work
img.show()
image.invert_colors()
img=QImage(image)
img.show()

Of course, this is not what you will actually be writing, it is an abstraction of the idea.

EDIT: In this case I would render the video (see this example & here), then take the image as a separate object, then render (again as a third object) with pyqt. To catch the location of the mouse click, look at this question, and finally, reference that point to the second object wich is the OpenCV image.