Simple Lossy Compression Image Format for Embedded Platforms

• Repository
• Online API Documents

• Simple.
• Lossy compression.
• A typical image compression ratio is somewhere between PNG and JPG.
• Small memory footprint, only a few hundred bytes of stack memory required for decoding.
• Designed for 16bpp color images and supports embedded-graphics; add features = ["embedded"] to Cargo.toml.
• Support for no_std, No alloc is needed for decoding.

• The latest version is recommended whenever possible.

• Example of a command line application that converts files in MPIC format and other formats such as PNG to each other

• Example of a GUI application that displays MPIC format files with embedded-graphics

• Online Image Viewer

• Divide the image into blocks of 8 x 8 pixels.
• Convert RGB with 8 bits per channel to YUV with 6 bits per channel.
• Thin out the U and V channels to 1/4.
• Because the color difference information is thinned out, even in the worst case, the compression is guaranteed to be more than half of the raw bitmap.
• Finally, lossless compression is performed using the sliding dictionary method.
• When decoding, these processes are performed in reverse order.

• All multi-byte data is encoded in little-endian.

• In version 0, only multiples of 8 are allowed for width and height.
• There is no limit to image size in version 1 or later.

• Image data follows the header.
• Image data is divided into 8 x 8 blocks and stored in chunks.
• If the image size is not a multiple of 8 x 8, the right and bottom edges are filled with a color interpolated from the surroundings to match a multiple of 8.
• Number of Chunks = ceil(width / 8) * ceil(height / 8)

• The first byte of each chunk of data indicates the data size, followed by the payload. The chunk size also indicates how the chunks are compressed.
• For uncompressed chunks, the data size (96), followed by the 64-byte Y channel, 16-byte U channel, and 16-byte V channel. 96 also serves as an identifier for uncompressed data. In practice, normal encoders do not use this mode.
• The Y channel stores all 8x8 data, while the U and V channels store only 4x4 pixels. The method of thinning the U and V channels is left to the encoder. The decoder should use nearest-neighbor interpolation to expand them by a factor of 2 in height and width.
• For a 6-bit compacted chunk, the data size is 72. The order of the data is the same as for the uncompressed chunk, but the 6 bits of the uncompressed chunk are compacted into 8 bits, so the data size is 3/4 of the uncompressed chunk.
• If the data size after compression exceeds 72 with other compression methods, the 6-bit compaction method shall be selected.

• RGB888 to YUV666

• YUV666 to RGB666 or RGB888

MIT

(C) 2023 Nerry