Files
esphome/esphome/components/image/__init__.py

767 lines
22 KiB
Python

from __future__ import annotations
import contextlib
from dataclasses import dataclass
import hashlib
import io
import logging
from pathlib import Path
import re
from PIL import Image, UnidentifiedImageError
from esphome import core, external_files
import esphome.codegen as cg
from esphome.components.const import CONF_BYTE_ORDER
import esphome.config_validation as cv
from esphome.const import (
CONF_DEFAULTS,
CONF_DITHER,
CONF_FILE,
CONF_ICON,
CONF_ID,
CONF_PATH,
CONF_RAW_DATA_ID,
CONF_RESIZE,
CONF_SOURCE,
CONF_TYPE,
CONF_URL,
)
from esphome.core import CORE, HexInt
_LOGGER = logging.getLogger(__name__)
DOMAIN = "image"
DEPENDENCIES = ["display"]
image_ns = cg.esphome_ns.namespace("image")
ImageType = image_ns.enum("ImageType")
@dataclass(frozen=True)
class ImageMetaData:
width: int
height: int
image_type: str
transparency: str
CONF_OPAQUE = "opaque"
CONF_CHROMA_KEY = "chroma_key"
CONF_ALPHA_CHANNEL = "alpha_channel"
CONF_INVERT_ALPHA = "invert_alpha"
CONF_IMAGES = "images"
KEY_METADATA = "metadata"
TRANSPARENCY_TYPES = (
CONF_OPAQUE,
CONF_CHROMA_KEY,
CONF_ALPHA_CHANNEL,
)
def get_image_type_enum(type):
return getattr(ImageType, f"IMAGE_TYPE_{type.upper()}")
def get_transparency_enum(transparency):
return getattr(TransparencyType, f"TRANSPARENCY_{transparency.upper()}")
class ImageEncoder:
"""
Superclass of image type encoders
"""
# Control which transparency options are available for a given type
allow_config = {CONF_ALPHA_CHANNEL, CONF_CHROMA_KEY, CONF_OPAQUE}
# All imageencoder types are valid
@staticmethod
def validate(value):
return value
def __init__(self, width, height, transparency, dither, invert_alpha):
"""
:param width: The image width in pixels
:param height: The image height in pixels
:param transparency: Transparency type
:param dither: Dither method
:param invert_alpha: True if the alpha channel should be inverted; for monochrome formats inverts the colours.
"""
self.transparency = transparency
self.width = width
self.height = height
self.data = [0 for _ in range(width * height)]
self.dither = dither
self.index = 0
self.invert_alpha = invert_alpha
self.path = ""
def convert(self, image, path):
"""
Convert the image format
:param image: Input image
:param path: Path to the image file
:return: converted image
"""
return image
def encode(self, pixel):
"""
Encode a single pixel
"""
def end_row(self):
"""
Marks the end of a pixel row
:return:
"""
@classmethod
def is_endian(cls) -> bool:
"""
Check if the image encoder supports endianness configuration
"""
return getattr(cls, "set_big_endian", None) is not None
@classmethod
def get_options(cls) -> list[str]:
"""
Get the available options for this image encoder
"""
options = [*OPTIONS]
if not cls.is_endian():
options.remove(CONF_BYTE_ORDER)
options.append(CONF_RAW_DATA_ID)
return options
def is_alpha_only(image: Image):
"""
Check if an image (assumed to be RGBA) is only alpha
"""
# Any alpha data?
if image.split()[-1].getextrema()[0] == 0xFF:
return False
return all(b.getextrema()[1] == 0 for b in image.split()[:-1])
class ImageBinary(ImageEncoder):
allow_config = {CONF_OPAQUE, CONF_INVERT_ALPHA, CONF_CHROMA_KEY}
def __init__(self, width, height, transparency, dither, invert_alpha):
self.width8 = (width + 7) // 8
super().__init__(self.width8, height, transparency, dither, invert_alpha)
self.bitno = 0
def convert(self, image, path):
if is_alpha_only(image):
image = image.split()[-1]
return image.convert("1", dither=self.dither)
def encode(self, pixel):
if self.invert_alpha:
pixel = not pixel
if pixel:
self.data[self.index] |= 0x80 >> (self.bitno % 8)
self.bitno += 1
if self.bitno == 8:
self.bitno = 0
self.index += 1
def end_row(self):
"""
Pad rows to a byte boundary
"""
if self.bitno != 0:
self.bitno = 0
self.index += 1
class ImageGrayscale(ImageEncoder):
allow_config = {CONF_ALPHA_CHANNEL, CONF_CHROMA_KEY, CONF_INVERT_ALPHA, CONF_OPAQUE}
def convert(self, image, path):
if is_alpha_only(image):
if self.transparency != CONF_ALPHA_CHANNEL:
_LOGGER.warning(
"Grayscale image %s is alpha only, but transparency is set to %s",
path,
self.transparency,
)
self.transparency = CONF_ALPHA_CHANNEL
image = image.split()[-1]
return image.convert("LA")
def encode(self, pixel):
b, a = pixel
if self.transparency == CONF_CHROMA_KEY:
if b == 1:
b = 0
if a != 0xFF:
b = 1
if self.invert_alpha:
b ^= 0xFF
if self.transparency == CONF_ALPHA_CHANNEL and a != 0xFF:
b = a
self.data[self.index] = b
self.index += 1
class ImageRGB565(ImageEncoder):
def __init__(self, width, height, transparency, dither, invert_alpha):
stride = 3 if transparency == CONF_ALPHA_CHANNEL else 2
super().__init__(
width * stride,
height,
transparency,
dither,
invert_alpha,
)
self.big_endian = True
def set_big_endian(self, big_endian: bool) -> None:
self.big_endian = big_endian
def convert(self, image, path):
return image.convert("RGBA")
def encode(self, pixel):
r, g, b, a = pixel
r = r >> 3
g = g >> 2
b = b >> 3
if self.transparency == CONF_CHROMA_KEY:
if r == 0 and g == 1 and b == 0:
g = 0
elif a < 128:
r = 0
g = 1
b = 0
rgb = (r << 11) | (g << 5) | b
if self.big_endian:
self.data[self.index] = rgb >> 8
self.index += 1
self.data[self.index] = rgb & 0xFF
self.index += 1
else:
self.data[self.index] = rgb & 0xFF
self.index += 1
self.data[self.index] = rgb >> 8
self.index += 1
if self.transparency == CONF_ALPHA_CHANNEL:
if self.invert_alpha:
a ^= 0xFF
self.data[self.index] = a
self.index += 1
class ImageRGB(ImageEncoder):
def __init__(self, width, height, transparency, dither, invert_alpha):
stride = 4 if transparency == CONF_ALPHA_CHANNEL else 3
super().__init__(
width * stride,
height,
transparency,
dither,
invert_alpha,
)
def convert(self, image, path):
return image.convert("RGBA")
def encode(self, pixel):
r, g, b, a = pixel
if self.transparency == CONF_CHROMA_KEY:
if r == 0 and g == 1 and b == 0:
g = 0
elif a < 128:
r = 0
g = 1
b = 0
self.data[self.index] = r
self.index += 1
self.data[self.index] = g
self.index += 1
self.data[self.index] = b
self.index += 1
if self.transparency == CONF_ALPHA_CHANNEL:
if self.invert_alpha:
a ^= 0xFF
self.data[self.index] = a
self.index += 1
class ReplaceWith:
"""
Placeholder class to provide feedback on deprecated features
"""
allow_config = {CONF_ALPHA_CHANNEL, CONF_CHROMA_KEY, CONF_OPAQUE}
def __init__(self, replace_with):
self.replace_with = replace_with
def validate(self, value):
raise cv.Invalid(
f"Image type {value} is removed; replace with {self.replace_with}"
)
IMAGE_TYPE = {
"BINARY": ImageBinary,
"GRAYSCALE": ImageGrayscale,
"RGB565": ImageRGB565,
"RGB": ImageRGB,
"TRANSPARENT_BINARY": ReplaceWith("'type: BINARY' and 'transparency: chroma_key'"),
"RGB24": ReplaceWith("'type: RGB'"),
"RGBA": ReplaceWith("'type: RGB' and 'transparency: alpha_channel'"),
}
TransparencyType = image_ns.enum("TransparencyType")
CONF_TRANSPARENCY = "transparency"
# If the MDI file cannot be downloaded within this time, abort.
IMAGE_DOWNLOAD_TIMEOUT = 30 # seconds
SOURCE_LOCAL = "local"
SOURCE_WEB = "web"
SOURCE_MDI = "mdi"
SOURCE_MDIL = "mdil"
SOURCE_MEMORY = "memory"
MDI_SOURCES = {
SOURCE_MDI: "https://raw.githubusercontent.com/Templarian/MaterialDesign/master/svg/",
SOURCE_MDIL: "https://raw.githubusercontent.com/Pictogrammers/MaterialDesignLight/refs/heads/master/svg/",
SOURCE_MEMORY: "https://raw.githubusercontent.com/Pictogrammers/Memory/refs/heads/main/src/svg/",
}
Image_ = image_ns.class_("Image")
INSTANCE_TYPE = Image_
def compute_local_image_path(value) -> Path:
url = value[CONF_URL] if isinstance(value, dict) else value
h = hashlib.new("sha256")
h.update(url.encode())
key = h.hexdigest()[:8]
base_dir = external_files.compute_local_file_dir(DOMAIN)
return base_dir / key
def local_path(value):
value = value[CONF_PATH] if isinstance(value, dict) else value
return str(CORE.relative_config_path(value))
def download_file(url, path):
external_files.download_content(url, path, IMAGE_DOWNLOAD_TIMEOUT)
return str(path)
def download_gh_svg(value, source):
mdi_id = value[CONF_ICON] if isinstance(value, dict) else value
base_dir = external_files.compute_local_file_dir(DOMAIN) / source
path = base_dir / f"{mdi_id}.svg"
url = MDI_SOURCES[source] + mdi_id + ".svg"
return download_file(url, path)
def download_image(value):
value = value[CONF_URL] if isinstance(value, dict) else value
return download_file(value, compute_local_image_path(value))
def is_svg_file(file):
if not file:
return False
with open(file, "rb") as f:
return "<svg" in str(f.read(1024))
def validate_file_shorthand(value):
value = cv.string_strict(value)
parts = value.strip().split(":")
if len(parts) == 2 and parts[0] in MDI_SOURCES:
match = re.match(r"^[a-zA-Z0-9\-]+$", parts[1])
if match is None:
raise cv.Invalid(f"Could not parse mdi icon name from '{value}'.")
return download_gh_svg(parts[1], parts[0])
if value.startswith("http://") or value.startswith("https://"):
return download_image(value)
value = cv.file_(value)
return local_path(value)
LOCAL_SCHEMA = cv.All(
{
cv.Required(CONF_PATH): cv.file_,
},
local_path,
)
def mdi_schema(source):
def validate_mdi(value):
return download_gh_svg(value, source)
return cv.All(
cv.Schema(
{
cv.Required(CONF_ICON): cv.string,
}
),
validate_mdi,
)
WEB_SCHEMA = cv.All(
{
cv.Required(CONF_URL): cv.string,
},
download_image,
)
TYPED_FILE_SCHEMA = cv.typed_schema(
{
SOURCE_LOCAL: LOCAL_SCHEMA,
SOURCE_WEB: WEB_SCHEMA,
}
| {source: mdi_schema(source) for source in MDI_SOURCES},
key=CONF_SOURCE,
)
def validate_transparency(choices=TRANSPARENCY_TYPES):
def validate(value):
if isinstance(value, bool):
value = str(value)
return cv.one_of(*choices, lower=True)(value)
return validate
def validate_type(image_types):
def validate(value):
value = cv.one_of(*image_types, upper=True)(value)
return IMAGE_TYPE[value].validate(value)
return validate
def validate_settings(value, path=()):
"""
Validate the settings for a single image configuration.
"""
conf_type = value[CONF_TYPE]
type_class = IMAGE_TYPE[conf_type]
transparency = value.get(CONF_TRANSPARENCY, CONF_OPAQUE).lower()
if transparency not in type_class.allow_config:
raise cv.Invalid(
f"Image format '{conf_type}' cannot have transparency: {transparency}"
)
invert_alpha = value.get(CONF_INVERT_ALPHA, False)
if (
invert_alpha
and transparency != CONF_ALPHA_CHANNEL
and CONF_INVERT_ALPHA not in type_class.allow_config
):
raise cv.Invalid("No alpha channel to invert")
if value.get(CONF_BYTE_ORDER) is not None and not type_class.is_endian():
raise cv.Invalid(
f"Image format '{conf_type}' does not support byte order configuration",
path=path,
)
if file := value.get(CONF_FILE):
file = Path(file)
if not is_svg_file(file):
try:
Image.open(file)
except UnidentifiedImageError as exc:
raise cv.Invalid(
f"File can't be opened as image: {file.absolute()}", path=path
) from exc
return value
IMAGE_ID_SCHEMA = {
cv.Required(CONF_ID): cv.declare_id(Image_),
cv.Required(CONF_FILE): cv.Any(validate_file_shorthand, TYPED_FILE_SCHEMA),
cv.GenerateID(CONF_RAW_DATA_ID): cv.declare_id(cg.uint8),
}
OPTIONS_SCHEMA = {
cv.Optional(CONF_RESIZE): cv.dimensions,
cv.Optional(CONF_DITHER, default="NONE"): cv.one_of(
"NONE", "FLOYDSTEINBERG", upper=True
),
cv.Optional(CONF_INVERT_ALPHA, default=False): cv.boolean,
cv.Optional(CONF_BYTE_ORDER): cv.one_of("BIG_ENDIAN", "LITTLE_ENDIAN", upper=True),
cv.Optional(CONF_TRANSPARENCY, default=CONF_OPAQUE): validate_transparency(),
}
DEFAULTS_SCHEMA = {
**OPTIONS_SCHEMA,
cv.Optional(CONF_TYPE): validate_type(IMAGE_TYPE),
}
OPTIONS = [key.schema for key in OPTIONS_SCHEMA]
# image schema with no defaults, used with `CONF_IMAGES` in the config
IMAGE_SCHEMA_NO_DEFAULTS = {
**IMAGE_ID_SCHEMA,
**{cv.Optional(key): OPTIONS_SCHEMA[key] for key in OPTIONS},
}
IMAGE_SCHEMA = cv.Schema(
{
**IMAGE_ID_SCHEMA,
**OPTIONS_SCHEMA,
cv.Required(CONF_TYPE): validate_type(IMAGE_TYPE),
}
)
def apply_defaults(image, defaults, path):
"""
Apply defaults to an image configuration
"""
type = image.get(CONF_TYPE, defaults.get(CONF_TYPE))
if type is None:
raise cv.Invalid(
"Type is required either in the image config or in the defaults", path=path
)
type_class = IMAGE_TYPE[type]
config = {
**{key: image.get(key, defaults.get(key)) for key in type_class.get_options()},
**{key.schema: image[key.schema] for key in IMAGE_ID_SCHEMA},
CONF_TYPE: image.get(CONF_TYPE, defaults.get(CONF_TYPE)),
}
validate_settings(config, path)
return config
def validate_defaults(value):
"""
Apply defaults to the images in the configuration and flatten to a single list.
"""
defaults = value[CONF_DEFAULTS]
result = []
# Apply defaults to the images: list and add the list entries to the result
for index, image in enumerate(value.get(CONF_IMAGES, [])):
result.append(apply_defaults(image, defaults, [CONF_IMAGES, index]))
# Apply defaults to images under the type keys and add them to the result
for image_type, type_config in value.items():
type_upper = image_type.upper()
if type_upper not in IMAGE_TYPE:
continue
type_class = IMAGE_TYPE[type_upper]
if isinstance(type_config, list):
# If the type is a list, apply defaults to each entry
for index, image in enumerate(type_config):
result.append(apply_defaults(image, defaults, [image_type, index]))
else:
# Handle transparency options for the type
for trans_type in set(type_class.allow_config).intersection(type_config):
for index, image in enumerate(type_config[trans_type]):
result.append(
apply_defaults(image, defaults, [image_type, trans_type, index])
)
return result
def typed_image_schema(image_type):
"""
Construct a schema for a specific image type, allowing transparency options
"""
return cv.Any(
cv.Schema(
{
cv.Optional(t.lower()): cv.ensure_list(
{
**IMAGE_ID_SCHEMA,
**{
cv.Optional(key): OPTIONS_SCHEMA[key]
for key in OPTIONS
if key != CONF_TRANSPARENCY
},
cv.Optional(
CONF_TRANSPARENCY, default=t
): validate_transparency((t,)),
cv.Optional(CONF_TYPE, default=image_type): validate_type(
(image_type,)
),
}
)
for t in IMAGE_TYPE[image_type].allow_config.intersection(
TRANSPARENCY_TYPES
)
}
),
# Allow a default configuration with no transparency preselected
cv.ensure_list(
{
**IMAGE_SCHEMA_NO_DEFAULTS,
cv.Optional(CONF_TYPE, default=image_type): validate_type(
(image_type,)
),
}
),
)
# The config schema can be a (possibly empty) single list of images,
# or a dictionary with optional keys `defaults:`, `images:` and the image types
def _config_schema(value):
if isinstance(value, list) or (
isinstance(value, dict) and (CONF_ID in value or CONF_FILE in value)
):
return cv.ensure_list(cv.All(IMAGE_SCHEMA, validate_settings))(value)
if not isinstance(value, dict):
raise cv.Invalid(
"Badly formed image configuration, expected a list or a dictionary",
)
return cv.All(
cv.Schema(
{
cv.Optional(CONF_DEFAULTS, default={}): DEFAULTS_SCHEMA,
cv.Optional(CONF_IMAGES, default=[]): cv.ensure_list(
{
**IMAGE_SCHEMA_NO_DEFAULTS,
cv.Optional(CONF_TYPE): validate_type(IMAGE_TYPE),
}
),
**{cv.Optional(t.lower()): typed_image_schema(t) for t in IMAGE_TYPE},
}
),
validate_defaults,
)(value)
CONFIG_SCHEMA = _config_schema
async def write_image(config, all_frames=False):
path = Path(config[CONF_FILE])
if not path.is_file():
raise core.EsphomeError(f"Could not load image file {path}")
resize = config.get(CONF_RESIZE)
try:
if is_svg_file(path):
import resvg_py
resize = resize or (None, None)
image_data = resvg_py.svg_to_bytes(
svg_path=str(path), width=resize[0], height=resize[1], dpi=100
)
# Convert bytes to Pillow Image
image = Image.open(io.BytesIO(image_data))
width, height = image.size
else:
image = Image.open(path)
width, height = image.size
if resize:
# Preserve aspect ratio
new_width_max = min(width, resize[0])
new_height_max = min(height, resize[1])
ratio = min(new_width_max / width, new_height_max / height)
width, height = int(width * ratio), int(height * ratio)
except (OSError, UnidentifiedImageError, ValueError) as exc:
raise core.EsphomeError(f"Could not read image file {path}: {exc}") from exc
if not resize and (width > 500 or height > 500):
_LOGGER.warning(
'The image "%s" you requested is very big. Please consider'
" using the resize parameter.",
path,
)
dither = (
Image.Dither.NONE
if config[CONF_DITHER] == "NONE"
else Image.Dither.FLOYDSTEINBERG
)
type = config[CONF_TYPE]
transparency = config.get(CONF_TRANSPARENCY, CONF_OPAQUE)
invert_alpha = config[CONF_INVERT_ALPHA]
frame_count = 1
if all_frames:
with contextlib.suppress(AttributeError):
frame_count = image.n_frames
if frame_count <= 1:
_LOGGER.warning("Image file %s has no animation frames", path)
total_rows = height * frame_count
encoder = IMAGE_TYPE[type](width, total_rows, transparency, dither, invert_alpha)
if byte_order := config.get(CONF_BYTE_ORDER):
# Check for valid type has already been done in validate_settings
encoder.set_big_endian(byte_order == "BIG_ENDIAN")
for frame_index in range(frame_count):
image.seek(frame_index)
pixels = encoder.convert(image.resize((width, height)), path).getdata()
for row in range(height):
for col in range(width):
encoder.encode(pixels[row * width + col])
encoder.end_row()
rhs = [HexInt(x) for x in encoder.data]
prog_arr = cg.progmem_array(config[CONF_RAW_DATA_ID], rhs)
image_type = get_image_type_enum(type)
trans_value = get_transparency_enum(encoder.transparency)
return prog_arr, width, height, image_type, trans_value, frame_count
async def _image_to_code(entry):
"""
Convert a single image entry to code and return its metadata.
:param entry: The config entry for the image.
:return: An ImageMetaData object
"""
prog_arr, width, height, image_type, trans_value, _ = await write_image(entry)
cg.new_Pvariable(entry[CONF_ID], prog_arr, width, height, image_type, trans_value)
return ImageMetaData(
width,
height,
entry[CONF_TYPE],
entry[CONF_TRANSPARENCY],
)
async def to_code(config):
cg.add_define("USE_IMAGE")
# By now the config will be a simple list.
# Use a subkey to allow for other data in the future
CORE.data[DOMAIN] = {
KEY_METADATA: {
entry[CONF_ID].id: await _image_to_code(entry) for entry in config
}
}
def get_all_image_metadata() -> dict[str, ImageMetaData]:
"""Get all image metadata."""
return CORE.data.get(DOMAIN, {}).get(KEY_METADATA, {})
def get_image_metadata(image_id: str) -> ImageMetaData | None:
"""Get image metadata by ID for use by other components."""
return get_all_image_metadata().get(image_id)