=====================
This task is simple and uses PIL (pillow) library. First step is mounting the google drive to the notebook where the dataset is present. Next, step is to read the path of the image data files using os library of python that handles the path the train dataset.
['germany-english-garden_10.jpg',
'taiwan-jiufen_15.jpg',
'japan-katsura-river_24.jpg',
'italy-garda-lake-sailing-club_18.jpg',
'irland-dingle_9.jpg',
'turkey-bodrum_5.jpg',
'germany-allgaeu-fliegenpilz_4.jpg',
'germany-garching-heide_7.jpg',
'england-london-bridge_2.jpg',
'DSCN0010_21.jpg']
PIL library provides a function named "getexif()" that provides the id's of all the Metadata tags and data that is present in the image.
img = Image.open(os.path.join(files,images))
meta_data_id = img.getexif()
By using a for loop, the image Metadata is mapped to the respective tags and stored in a dictionary in the given format {tag_name: data}.
tag = TAGS.get(_id,_id)
meta_data = meta_data_id.get(_id)
Important is to naote that the Metadata present in 'bytes' format needs to be decoded. To encounter this, try and except are used to ignore the values that are unable to decode and replace such values with an empty string.
try:
if isinstance(meta_data,bytes):
meta_data = meta_data.decode()
except:
meta_data = ""
=====================
First, a dictionary is formed with teh format {imagename : id} using the below function ehcih is self explanatory.
id_image = {}
for image in os.listdir(files):
id = image.split("_")[-1].split(".")[0]
id_image[image] = int(id)
So, using the basic ".spit()" function of string split the image file name about '_' and then splitting the string about '.'. This is explained below
name = ['germany-english-garden_10.jpg']
name = ['germany-english-garden','10.jpeg'] (after splitting about "_" and taking string at index 1)
name = ['10','jpeg'] (after splitting about "." and taking string at index 0)
Second, a Pandas Dataframe is initialized with all 0(zero) values,index range(0-24), columns=( Imagename , Resized Image, Rotated Image, MetadataInfo). The index is provided on the bases of the id given in the image file name as imagename_id.jpeg.
df = pd.DataFrame(0,index=np.arange(0,max(id_image.values())+1),columns=['Image_name','Resized Image','Rotated Image','MetadataInfo'])
df.style.set_properties(subset=["Image_name"], **{'text-align': 'center'})
df.head()
As, provided the names and metadata associated with the image file is inserted into the dataframe at their respective id. The metadata dictionary which is formed in the task-1 is used for metadata insertion in the dataframe. The values (tag : metadata) of respective keys are concatenated into a single string with a line break using "\n".join().
for image, id in id_image.items():
df['Image_name'][id] = image
meta_tokens = {}
for key, value in data.items():
ls = data[key]
temp = []
for tag in ls:
temp.append(str(tag[0])+": "+str(tag[1]))
meta_tokens[key] = temp
metadatainfo = {}
for keys,values in meta_tokens.items():
metadatainfo[keys] = "\n".join(values)
for keys, values in metadatainfo.items():
df['MetadataInfo'][df['Image_name'] == keys] = values
Now he important part, rendering or inserting the image into pandas
dataframe. This is done using HTML "" tag.
The below functions return the thumbail of the image present at the provided location. The image is resized into (512x512) and returned as the thumbnail. The if statement hanndles the indices that are not defined ins the dataset of the image id.
def get_thumbnail_unrotated(path):
if path == "":
return ""
i = Image.open(path)
i = i.resize(size=(512,512))
i.thumbnail((512, 512), Image.LANCZOS)
return i
This function is same as above, just modified as it rotates the image 90 degree in clockwise direction.
def get_thumbnail_rotated(path):
if path == "":
return ""
i = Image.open(path)
i = i.resize(size=(512,512))
i = i.rotate(angle=-90)
i.thumbnail((512, 512), Image.LANCZOS)
return i
Thes functions are the HTML formatters that return the tag
"" with image path as the src location.
def image_formatter_unrotated(im):
if im == "":
return 0
return f'<img src="data:image/jpeg;base64,{image_base64_unrotated(im)}">'
def image_formatter_rotated(im):
if im == "":
return 0
return f'<img src="data:image/jpeg;base64,{image_base64_rotate(im)}">'
These functions check for the path, if its string then they convert the path into a thumbnail, otherwise it writes data in memory buffer and decodes the buffer using base64encode (utf-8)
def image_base64_unrotated(im):
if isinstance(im, str):
im = get_thumbnail_unrotated(im)
with BytesIO() as buffer:
im.save(buffer, 'jpeg')
return base64.b64encode(buffer.getvalue()).decode()
def image_base64_rotate(im):
if isinstance(im, str):
im = get_thumbnail_rotated(im)
with BytesIO() as buffer:
im.save(buffer, 'jpeg')
return base64.b64encode(buffer.getvalue()).decode()
A new column a inserted into the dataframe that contains the image file path.
# Saving the file path of the images
df['file_1'] = df.Image_name.map(lambda id: contradiction(id))
# converting the path of the image to the HTML tag for rotated and
# unrotated image
df['Resized Image'] = df.file_1.map(lambda f: get_thumbnail_unrotated(f))
df['Rotated Image'] = df.file_1.map(lambda f: get_thumbnail_rotated(f))
df.head()
Now, the dataframe is completed using following function that converts
the "" tag to the image.
HTML(df[["Image_name","Resized Image","Rotated Image",'MetadataInfo']].to_html(formatters={'Resized Image': image_formatter_unrotated, 'Rotated Image':image_formatter_rotated}, escape=False))
