Segment Image
We will use this function to visualize the results.
Python
#IMAGE_DIR = 'Mask_RCNN/images/' # Load a random image from the images folder #file_names = next(os.walk(IMAGE_DIR))[2] #image = skimage.io.imread(os.path.join(IMAGE_DIR, np.random.choice(file_names))) image = skimage.io.imread(" / content / image.jpg") plt.imshow(image) plt.title( 'Original' ) plt.axis( 'off' ) plt.show() # Run detection results = model.detect([image], verbose = 1 ) # Visualize results r = results[ 0 ] visualize.display_instances(image, r[ 'rois' ], r[ 'masks' ], r[ 'class_ids' ], class_names, r[ 'scores' ]) |
Output:
Processing 1 images
image shape: (1282, 1920, 3) min: 0.00000 max: 255.00000 uint8
molded_images shape: (1, 1024, 1024, 3) min: -123.70000 max: 134.10000 float64
image_metas shape: (1, 93) min: 0.00000 max: 1920.00000 float64
anchors shape: (1, 261888, 4) min: -0.35390 max: 1.29134 float32
Python
# For converting roi coordinates generated using Mask R-CNN # to (x1, y1, width, height) def convertCoords(coords): # Input : (y1, x1, y2, x2) y1, x1, y2, x2 = coords width,height = x2 - x1, y2 - y1 return (x1, y1, width, height) # Extracting masks and rois from the result def separateEntities(r): masks = [r[ 'masks' ][:,:,i].astype("uint8") for i in range ( len (r[ 'class_ids' ]))] rects = [(convertCoords(roi)) for roi in r[ 'rois' ]] return masks,rects # Generate masks masks, rects = separateEntities(r) |
Visualizing picture of masks generated using Mask R-CNN
Python3
# Create subplots with 1 row and 2 columns fig, (ax1, ax2) = plt.subplots( 1 , 2 , figsize = ( 10 , 5 )) # Display the first mask on the left subplot ax1.imshow(masks[ 0 ]) ax1.axis( 'off' ) # Display the second mask on the right subplot ax2.imshow(masks[ 1 ]) ax2.axis( 'off' ) # Show the subplots plt.show() |
Output:
Visualizing Segmented Image using Mask R-CNN
Python3
# Create subplots with 1 row and 2 columns fig, (ax1, ax2) = plt.subplots( 1 , 2 , figsize = ( 10 , 5 )) # Display the first mask on the left subplot ax1.imshow(cv2.bitwise_and(image,image,mask = (r[ 'masks' ][:,:, 0 ].astype("uint8")))) ax1.axis( 'off' ) # Display the second mask on the right subplot ax2.imshow(cv2.bitwise_and(image,image,mask = (r[ 'masks' ][:,:, 1 ].astype("uint8")))) ax2.axis( 'off' ) # Show the subplots plt.show() |
Output:
The masks generated using Mask-RCNN are not precise, there are visible background details, we will use GrabCut to remove the undesired background by refining the masks.
Applying GrabCut
GrabCut is available in OpenCV as cv2.grabCut(img, mask, rect, bgdModel, fgdModel, iterCount, mode)
Let’s explore the arguments:
- img: This argument represents the input image on which we want to perform the GrabCut algorithm.
- mask: The mask image is used to define the regions of the image as background, foreground, probable background/foreground, etc. This is achieved by assigning specific flags to different areas of the mask image. The flags used are
- cv.GC_BGD (background),
- cv.GC_FGD (foreground),
- cv.GC_PR_BGD (probable background),
- cv.GC_PR_FGD (probable foreground), or you can directly pass 0, 1, 2, or 3 to represent these regions in the image.
- rect: This argument represents the coordinates of a rectangle that encloses the foreground object in the format (x, y, w, h). The rectangle is used to initialize the GrabCut algorithm and provide an initial estimate of the foreground and background regions.
- bdgModel, fgdModel: These are numpy arrays that are internally used by the GrabCut algorithm. You need to create two zero arrays of type np.float64 and size (1, 65) to store the model parameters.
- iterCount: The iterCount specifies the number of iterations the GrabCut algorithm should run. More iterations can lead to better segmentation results, but at the cost of increased computation time.
- mode: The mode parameter determines whether we are providing the rectangle coordinates (cv.GC_INIT_WITH_RECT), mask (cv.GC_INIT_WITH_MASK), or a combination of both. This choice determines whether we are initially drawing a rectangle around the object of interest or providing additional touch-up strokes later.
Python
def applyGrabCut(image,mask,rect,iters): fgModel = np.zeros(( 1 , 65 ), dtype = " float ") bgModel = np.zeros(( 1 , 65 ), dtype = " float ") # apply GrabCut using the the bounding box segmentation method (mask_grab, bgModel, fgModel) = cv2.grabCut(image, mask, rect, bgModel, fgModel, iterCount = iters, mode = cv2.GC_INIT_WITH_RECT) values = ( ("Definite Background", cv2.GC_BGD), ("Probable Background", cv2.GC_PR_BGD), ("Definite Foreground", cv2.GC_FGD), ("Probable Foreground", cv2.GC_PR_FGD), ) valueMasks = {} for name,value in values: valueMasks[name] = (mask_grab = = cv2.GC_PR_FGD).astype("uint8") * 255 return valueMasks |
Generating Picture of Masks and Segmented image
Python
# Create subplots with 1 row and 2 columns fig, axs = plt.subplots( 2 , 2 , figsize = ( 11 , 7 )) # Process the first mask and display the result vm1 = applyGrabCut(image, masks[ 0 ], rects[ 0 ], 10 ) mother = cv2.bitwise_and(image, image, mask = vm1[ 'Definite Foreground' ]) axs[ 0 , 0 ].imshow(vm1[ 'Probable Foreground' ]) axs[ 0 , 0 ].set_title( 'Mother' ) axs[ 0 , 0 ].axis( 'off' ) # Display the processed first mask axs[ 0 , 1 ].imshow(mother) axs[ 0 , 1 ].set_title( 'Mother' ) axs[ 0 , 1 ].axis( 'off' ) # Process the second mask and display the result vm2 = applyGrabCut(image, masks[ 1 ], rects[ 1 ], 10 ) child = cv2.bitwise_and(image, image, mask = vm2[ 'Definite Foreground' ]) axs[ 1 , 0 ].imshow(vm2[ 'Probable Foreground' ]) axs[ 1 , 0 ].set_title( 'Child' ) axs[ 1 , 0 ].axis( 'off' ) # Display the processed second mask axs[ 1 , 1 ].imshow(child) axs[ 1 , 1 ].set_title( 'Child' ) axs[ 1 , 1 ].axis( 'off' ) # Show the subplots plt.show() |
Output:
Combining both masks
Python
# Create subplots with 1 row and 2 columns fig, (ax1, ax2) = plt.subplots( 2 , 1 , figsize = ( 8 , 8 )) # Display the combined mask combined_mask = vm1[ 'Definite Foreground' ] | vm2[ 'Definite Foreground' ] ax1.imshow(combined_mask) ax1.axis( 'off' ) ax1.set_title( 'Combined Mask' ) # Display the second segmented image result = mother | child ax2.imshow(result) ax2.axis( 'off' ) ax2.set_title( 'Segmented IMage' ) # Show the subplots plt.show() |
Output:
You can run this implementation on Google Colab Notebook
Image Segmentation with Mask R-CNN, GrabCut, and OpenCV
Image segmentation plays a crucial role in computer vision tasks, enabling machines to understand and analyze visual content at a pixel level. It involves dividing an image into distinct regions or objects, facilitating object recognition, tracking, and scene understanding. In this article, we explore three popular image segmentation techniques: Mask R-CNN, GrabCut, and OpenCV.
Let’s understand, What Image Segmentation with Mask R-CNN and GrabCut are?
What is R-CNN?
R-CNN stands for Region-based Convolutional Neural Network. It is a ground-breaking object detection system that combines object localization and recognition into an end-to-end deep learning framework.
RNN can be summarised in the following ways.
- Region Proposal: Initially, a region proposal algorithm (such as selective search) generates a set of potential bounding box regions in an image that are likely to contain objects of interest. These regions serve as candidate object locations.
- Feature Extraction: Each region proposal is then individually cropped and resized to a fixed size and passed through a pre-trained CNN (such as AlexNet or VGGNet). The CNN extracts high-level features from the region, transforming it into a fixed-length feature vector.
- Classification and Localization: The feature vector obtained from the CNN is fed into separate fully connected layers. The classification layer predicts the probability of different object classes present in the region, while the regression layer refines the coordinates of the bounding box, improving localization accuracy.
- Non-Maximum Suppression (NMS): To eliminate redundant detections, non-maximum suppression is applied. It removes overlapping bounding boxes, keeping only the most confident detection for each object instance.