Image Utilities Examples

Examples on using functions from toolbox.im_utils¶

this notebook offers some visual examples on how some of the functions in toolbox.im_utils can be used. For the complete and extensive description on all the image utility functions, please check out the Reference Page

In [1]:

from toolbox.im_utils import *
im = get_im('https://picsum.photos/id/177/500/300')
person_bbox = {'x0': 227, 'y0': 152, 'x1': 284, 'y1': 298}

from toolbox.im_utils import * im = get_im('https://picsum.photos/id/177/500/300') person_bbox = {'x0': 227, 'y0': 152, 'x1': 284, 'y1': 298}

Basic Image Processing Functions¶

We can get images of different size

In [2]:

get_pil_im(im)

get_pil_im(im)

Out[2]:

In [3]:

get_pil_im(im_max_long_edge(im, 300))

get_pil_im(im_max_long_edge(im, 300))

Out[3]:

In [4]:

get_pil_im(im_min_short_edge(im, 700))

get_pil_im(im_min_short_edge(im, 700))

Out[4]:

Cropping an Image¶

In [5]:

im_person = get_pil_im(im_crop(im, **person_bbox))
print(im_person.size); im_person

im_person = get_pil_im(im_crop(im, **person_bbox)) print(im_person.size); im_person

(57, 146)

Out[5]:

Padding an Image¶

In [6]:

get_pil_im(
    im_centre_pad(im_person, (150,150))
)

get_pil_im( im_centre_pad(im_person, (150,150)) )

Out[6]:

Concatenating two images in different flavours¶

Vertically...

In [7]:

get_pil_im(
    image_stack('https://picsum.photos/id/20/400/300', 'https://picsum.photos/id/26/400/350',
               do_vstack = True, black_line_thickness= 2)
)

get_pil_im( image_stack('https://picsum.photos/id/20/400/300', 'https://picsum.photos/id/26/400/350', do_vstack = True, black_line_thickness= 2) )

Out[7]:

Horizontally...

In [8]:

get_pil_im(
    image_stack('https://picsum.photos/id/20/200/300', 'https://picsum.photos/id/26/400/300',
               do_vstack = False, black_line_thickness= 2)
)

get_pil_im( image_stack('https://picsum.photos/id/20/200/300', 'https://picsum.photos/id/26/400/300', do_vstack = False, black_line_thickness= 2) )

Out[8]:

or splitting according to a percentage...

In [9]:

get_pil_im(
    image_stack('https://picsum.photos/id/20/400/300', 'https://picsum.photos/id/26/400/300',
                split_pct = 0.5,
                do_vstack = False, black_line_thickness= 2)
)

get_pil_im( image_stack('https://picsum.photos/id/20/400/300', 'https://picsum.photos/id/26/400/300', split_pct = 0.5, do_vstack = False, black_line_thickness= 2) )

Out[9]:

Annotation Functions¶

writing text¶

In [10]:

get_pil_im(im_write_text(im, "Hello World!", x = 200, y = 10, tup_font_rgb = (250,0,0)))

get_pil_im(im_write_text(im, "Hello World!", x = 200, y = 10, tup_font_rgb = (250,0,0)))

Out[10]:

Drawing Bounding Box¶

you can do so simply using PIL and can even get a label for your bounding box!

In [11]:

get_pil_im(
    im_draw_bbox(im, **person_bbox, color = 'white', width = 1, caption = 'person')
)

get_pil_im( im_draw_bbox(im, **person_bbox, color = 'white', width = 1, caption = 'person') )

Out[11]:

or you can use the bbox_visualizer option which draws more visually pleasing bounding boxes

In [12]:

get_pil_im(
    im_draw_bbox(im, **person_bbox, color = 'white', width = 1, caption = 'person', use_bbv = True)
)

get_pil_im( im_draw_bbox(im, **person_bbox, color = 'white', width = 1, caption = 'person', use_bbv = True) )

Out[12]:

Masking a bounding box¶

you can mask out the area outside the bounding box with your color of choice

In [13]:

get_pil_im(
    im_mask_bbox(im, l_bboxes = [person_bbox], mask_rgb_tup= (250,0,0))
)

get_pil_im( im_mask_bbox(im, l_bboxes = [person_bbox], mask_rgb_tup= (250,0,0)) )

Out[13]:

or you can mask out the bounding box itself... which could be useful in a privacy setting!

In [14]:

get_pil_im(
    im_mask_bbox(im, l_bboxes = [person_bbox], b_mask_bg = False)
)

get_pil_im( im_mask_bbox(im, l_bboxes = [person_bbox], b_mask_bg = False) )

Out[14]:

Mask-related Functions¶

In [15]:

import numpy as np
person_mask = np.zeros(im.shape[:2], dtype = np.uint8)
person_mask[152:298 ,227:284] = 1

import numpy as np person_mask = np.zeros(im.shape[:2], dtype = np.uint8) person_mask[152:298 ,227:284] = 1

you can apply an color to a mask...

In [16]:

get_pil_im(
    im_color_mask(im, mask_array = person_mask, alpha = .5)
)

get_pil_im( im_color_mask(im, mask_array = person_mask, alpha = .5) )

Out[16]:

to color the inverse of the mask you can do something like this...

In [17]:

get_pil_im(
    im_color_mask(im, mask_array = np.logical_not(person_mask).astype('uint8'), alpha = 0.7)
)

get_pil_im( im_color_mask(im, mask_array = np.logical_not(person_mask).astype('uint8'), alpha = 0.7) )

Out[17]:

to do background removal given a mask...

In [18]:

get_pil_im(
    im_apply_mask(im, mask_array = person_mask, bg_rgb_tup = [50,205,50])
)

get_pil_im( im_apply_mask(im, mask_array = person_mask, bg_rgb_tup = [50,205,50]) )

Out[18]:

alternatively you can blur out the background

In [19]:

get_pil_im(
    im_apply_mask(im, mask_array = person_mask, bg_blur_radius = 5)
)

get_pil_im( im_apply_mask(im, mask_array = person_mask, bg_blur_radius = 5) )

Out[19]:

or greyscale it out

In [20]:

get_pil_im(
    im_apply_mask(im, mask_array = person_mask, bg_greyscale = True)
)

get_pil_im( im_apply_mask(im, mask_array = person_mask, bg_greyscale = True) )

Out[20]:

and if your mask's edges are too shape, you can soften it too!

In [21]:

get_pil_im(
    im_apply_mask(im, mask_array = person_mask, mask_gblur_radius= 5, bg_greyscale = True)
)

get_pil_im( im_apply_mask(im, mask_array = person_mask, mask_gblur_radius= 5, bg_greyscale = True) )

Out[21]: