Concepts

The Python Imaging Library handles raster images — rectangles of pixel data.

Bands

An image can consist of one or more bands of data. Pillow allows you to store several bands in a single image, provided they all have the same dimensions and depth. For example, a PNG image might have 'R', 'G', 'B', and 'A' bands.

Working with Bands

im = Image.open("photo.png") print(im.getbands()) # ('R', 'G', 'B', 'A') # Split into individual band images r, g, b, a = im.split() # Merge bands back (maybe with a modified alpha) out = Image.merge("RGBA", (r, g, b, a))

Modes

The mode of an image defines the type and depth of a pixel. Each pixel uses the full range of the bit depth.

Mode	Description	Bands
`1`	1-bit pixels, black and white. Stored with one pixel per byte.	1
`L`	8-bit pixels, grayscale (Luminance).	1
`P`	8-bit pixels, mapped to any other mode using a colour palette.	1
`RGB`	3×8-bit pixels, true colour.	3
`RGBA`	4×8-bit pixels, true colour with transparency mask.	4
`CMYK`	4×8-bit pixels, colour separation. Used in print workflows.	4
`YCbCr`	3×8-bit pixels, colour video format (JPEG standard).	3
`LAB`	3×8-bit pixels, the Lab color space.	3
`HSV`	3×8-bit pixels, Hue (0-255 scaled 0-360°), Saturation, Value.	3
`I`	32-bit signed integer pixels.	1
`F`	32-bit floating point pixels.	1
`LA`	L with alpha.	2
`PA`	P with alpha.	2
`RGBa`	RGB with pre-multiplied alpha.	4
`La`	L with pre-multiplied alpha.	2
`I;16`	16-bit unsigned integer pixels.	1
`I;16B`	16-bit big-endian unsigned integer pixels.	1

Mode Conversion

# Check current mode print(im.mode) # 'RGB' # Convert to grayscale gray = im.convert("L") # Grayscale → RGB (for applying colour operations) rgb = gray.convert("RGB") # RGB → CMYK (for print) cmyk = im.convert("CMYK")

Coordinate System

Pillow uses a Cartesian coordinate system with the origin (0,0) in the upper left corner. Note that coordinates refer to pixel corners, not pixel centers. So the bounding box for a 100×100 pixel image is: (0, 0, 100, 100).

Concept	Value
Origin	Top-left corner `(0, 0)`
X-axis	Increases rightward (width)
Y-axis	Increases downward (height)
Bounding box	`(left, upper, right, lower)` — exclusive on right and lower

Palette Mode

Palette (P mode) images use a colour lookup table. Each pixel value is an index into this table. This is used by GIF, which is limited to 256 unique colours per frame.

Palette Operations

# Convert to palette mode — quantize to 256 colors p_img = im.convert("P") # Quantize with explicit number of colors p_img = im.quantize(colors=128) # Get the raw palette data palette = p_img.getpalette() # flat list: [R,G,B, R,G,B, ...] # Save as GIF p_img.save("output.gif")

The Info Attribute

The im.info dictionary is populated by the file decoder with format-specific metadata. Common entries include DPI, comments, and EXIF data. This dictionary is not automatically written on save() — you must explicitly pass parameters to the save method.

Reading Image Metadata

im = Image.open("photo.jpg") # Access info dict print(im.info) # {'jfif': 257, 'jfif_version': (1, 1), 'dpi': (72, 72), ...} # EXIF via getexif() from PIL import ExifTags exif = im.getexif() for tag_id, val in exif.items(): print(ExifTags.TAGS.get(tag_id, tag_id), val)

Filters

The ImageFilter module defines a set of predefined image filters and a generic kernel filter. Filters are applied using im.filter().

Filter	Category	Description
`BLUR`	Blur	Box blur.
`GaussianBlur(radius)`	Blur	Gaussian bell-curve blur. Default radius=2.
`BoxBlur(radius)`	Blur	Each output pixel is the average of the square neighbourhood.
`CONTOUR`	Edge	Find image contours.
`DETAIL`	Sharpen	Enhance detail.
`EDGE_ENHANCE`	Edge	Edge enhancement filter.
`EDGE_ENHANCE_MORE`	Edge	More aggressive edge enhancement.
`EMBOSS`	Effect	Emboss image.
`FIND_EDGES`	Edge	Find sharp edges.
`SHARPEN`	Sharpen	Sharpen image.
`SMOOTH`	Smooth	Smooth image.
`SMOOTH_MORE`	Smooth	More aggressive smoothing.
`UnsharpMask(radius, percent, threshold)`	Sharpen	Unsharp mask. radius=2, percent=150, threshold=3.
`Kernel(size, kernel)`	Custom	Create custom convolution kernel.
`RankFilter(size, rank)`	Rank	Picks rank-th minimum pixel in each neighbourhood.
`MedianFilter(size)`	Rank	Picks median pixel — good for "salt and pepper" noise.
`MinFilter(size)`	Rank	Picks minimum (darkest) pixel in neighbourhood.
`MaxFilter(size)`	Rank	Picks maximum (brightest) pixel in neighbourhood.