Merge pull request #386 from aarshivv/palette-cielab-transparency-lzw

support palette expansion, CIE L*a*b*, transparency mask, and lzw end…

src/decoder/image.rs

@@ -597,20 +597,25 @@ impl Image {
                     ),
                 )),
             },
-            // Unsupported due to inherently heterogeneous sample types. This is represented as:
-            // ```
-            // struct TiffLab /* Interpretation 8* {
-            //     pub L: u8, // SampleFormat::Uint
-            //     pub a: i8, // SampleFormat::Int
-            //     pub b: i8, // SampleFormat::Int
-            // }
-            // ```
-            PhotometricInterpretation::CIELab => Err(TiffError::UnsupportedError(
-                TiffUnsupportedError::InterpretationWithBits(
-                    PhotometricInterpretation::CIELab,
-                    vec![self.bits_per_sample; self.samples as usize],
-                ),
-            )),
+            // CIE L*a*b* (TIFF6 Section 23, pp 110-115) encodes a/b as signed int8
+            // (-128..=127) while ICC L*a*b* (photometric=9) uses unsigned samples with
+            // a +128 bias. For 8-bit we re-bias a/b in post-processing (see
+            // `cielab_to_icclab`) so callers see the same encoding as ICCLab.
+            //
+            // Only 8-bit is accepted here. 16-bit CIELab is a valid format that libtiff
+            // does decode (`putcontig8bitCIELab16`, selected by `initCIELabConversion`
+            // in tif_getimage.c line 2216 @ 04884181e8903915038be553bfab404302d98ea0),
+            // but it uses L=uint16 with a/b=int16 — a mixed layout that `SampleFormat`,
+            // which applies uniformly to every sample, cannot express. Left as follow-up.
+            PhotometricInterpretation::CIELab => match self.photometric_samples {
+                3 if matches!(self.bits_per_sample, 8) => Ok(ColorType::Lab(self.bits_per_sample)),
+                _ => Err(TiffError::UnsupportedError(
+                    TiffUnsupportedError::InterpretationWithBits(
+                        PhotometricInterpretation::CIELab,
+                        vec![self.bits_per_sample; self.samples as usize],
+                    ),
+                )),
+            },
             PhotometricInterpretation::CIELogL => match self.samples {
                 1 if matches!(self.bits_per_sample, 32)
                     && matches!(self.sample_format, SampleFormat::IEEEFP) =>
@@ -666,14 +671,16 @@ impl Image {
                 ),
             )),
             PhotometricInterpretation::RGBPalette => Ok(ColorType::Palette(self.bits_per_sample)),
-            PhotometricInterpretation::TransparencyMask | PhotometricInterpretation::Unknown(_) => {
-                Err(TiffError::UnsupportedError(
-                    TiffUnsupportedError::InterpretationWithBits(
-                        self.photometric_interpretation,
-                        vec![self.bits_per_sample; self.samples as usize],
-                    ),
-                ))
+            // TransparencyMask is a single-channel mask, treat as grayscale.
+            PhotometricInterpretation::TransparencyMask => {
+                Ok(ColorType::Gray(self.bits_per_sample))
             }
+            PhotometricInterpretation::Unknown(_) => Err(TiffError::UnsupportedError(
+                TiffUnsupportedError::InterpretationWithBits(
+                    self.photometric_interpretation,
+                    vec![self.bits_per_sample; self.samples as usize],
+                ),
+            )),
         }
     }
 
@@ -1046,6 +1053,7 @@ impl Image {
             | ColorType::YCbCr(n)
             | ColorType::Gray(n)
             | ColorType::Palette(n)
+            | ColorType::Lab(n)
             | ColorType::Multiband {
                 bit_depth: n,
                 num_samples: _,
@@ -1057,6 +1065,7 @@ impl Image {
             | ColorType::YCbCr(n)
             | ColorType::Gray(n)
             | ColorType::Palette(n)
+            | ColorType::Lab(n)
             | ColorType::Multiband {
                 bit_depth: n,
                 num_samples: _,
@@ -1214,9 +1223,12 @@ impl Image {
                 // the row is exact).
                 super::mask_samples_to_bit_depth(out_row, tiff_bps, out_bps, self.sample_format);
 
-                if photometric_interpretation == PhotometricInterpretation::WhiteIsZero {
-                    super::invert_colors(out_row, color_type, self.sample_format)?;
-                }
+                super::post_process_row(
+                    out_row,
+                    photometric_interpretation,
+                    color_type,
+                    self.sample_format,
+                )?;
             }
         } else if is_output_chunk_rows && is_all_bits {
             // Here we can read directly into the output buffer itself.
@@ -1244,9 +1256,12 @@ impl Image {
                 self.sample_format,
             );
 
-            if photometric_interpretation == PhotometricInterpretation::WhiteIsZero {
-                super::invert_colors(tile, color_type, self.sample_format)?;
-            }
+            super::post_process_row(
+                tile,
+                photometric_interpretation,
+                color_type,
+                self.sample_format,
+            )?;
         } else if chunk_row_bytes > data_row_bytes && self.predictor == Predictor::FloatingPoint {
             // The floating point predictor shuffles the padding bytes into the encoded output, so
             // this case is handled specially when needed.
@@ -1272,9 +1287,12 @@ impl Image {
                 }
                 // No masking here: the floating point predictor requires
                 // IEEEFP samples, which are never depth-promoted.
-                if photometric_interpretation == PhotometricInterpretation::WhiteIsZero {
-                    super::invert_colors(row, color_type, self.sample_format)?;
-                }
+                super::post_process_row(
+                    row,
+                    photometric_interpretation,
+                    color_type,
+                    self.sample_format,
+                )?;
             }
         } else if is_all_bits {
             // We read row-by-row but each row fits in its output buffer.
@@ -1311,9 +1329,12 @@ impl Image {
                     self.sample_format,
                 );
 
-                if photometric_interpretation == PhotometricInterpretation::WhiteIsZero {
-                    super::invert_colors(row, color_type, self.sample_format)?;
-                }
+                super::post_process_row(
+                    row,
+                    photometric_interpretation,
+                    color_type,
+                    self.sample_format,
+                )?;
             }
         } else {
             // The encoded data potentially takes up more space than the output data so we must be
@@ -1363,9 +1384,12 @@ impl Image {
                     self.sample_format,
                 );
 
-                if photometric_interpretation == PhotometricInterpretation::WhiteIsZero {
-                    super::invert_colors(row, color_type, self.sample_format)?;
-                }
+                super::post_process_row(
+                    row,
+                    photometric_interpretation,
+                    color_type,
+                    self.sample_format,
+                )?;
             }
         }
 

src/decoder/mod.rs

@@ -3,8 +3,8 @@ use std::io::{self, Read, Seek};
 use std::num::NonZeroUsize;
 
 use crate::tags::{
-    ByteOrder, IfdPointer, PlanarConfiguration, Predictor, SampleFormat, Tag, TiffVariant, Type,
-    ValueBuffer,
+    ByteOrder, IfdPointer, PhotometricInterpretation, PlanarConfiguration, Predictor, SampleFormat,
+    Tag, TiffVariant, Type, ValueBuffer,
 };
 use crate::{
     ColorType, Directory, TiffError, TiffFormatError, TiffResult, TiffUnsupportedError, UsageError,
@@ -711,6 +711,40 @@ fn invert_colors(
     Ok(())
 }
 
+/// Re-bias CIE L*a*b* a/b channels from signed to unsigned encoding.
+///
+/// CIELab (photometric=8) stores a/b as int8 where 0 is neutral.
+/// ICCLab (photometric=9) stores them as uint8 where 128 is neutral.
+/// Adding 128 (wrapping) converts between the two representations.
+/// See libtiff's `TIFFCIELabToXYZ` for reference (tif_color.c line 45
+/// @ 04884181e8903915038be553bfab404302d98ea0).
+fn cielab_to_icclab(buf: &mut [u8]) {
+    for [_, a, b] in buf.as_chunks_mut::<3>().0 {
+        *a = a.wrapping_add(128);
+        *b = b.wrapping_add(128);
+    }
+}
+
+/// Apply photometric-interpretation-specific post-processing to a decoded row.
+///
+/// Some interpretations need a fixup once the raw samples are in the output
+/// buffer: `WhiteIsZero` inverts intensities, and `CIELab` re-biases the signed
+/// a/b channels into the unsigned ICCLab encoding. These are mutually exclusive,
+/// so at most one applies to any given row.
+fn post_process_row(
+    buf: &mut [u8],
+    photometric_interpretation: PhotometricInterpretation,
+    color_type: ColorType,
+    sample_format: SampleFormat,
+) -> TiffResult<()> {
+    match photometric_interpretation {
+        PhotometricInterpretation::WhiteIsZero => invert_colors(buf, color_type, sample_format)?,
+        PhotometricInterpretation::CIELab => cielab_to_icclab(buf),
+        _ => {}
+    }
+    Ok(())
+}
+
 /// Fix endianness. If `byte_order` matches the host, then conversion is a no-op.
 fn fix_endianness(buf: &mut [u8], byte_order: ByteOrder, bit_depth: u8) {
     let host = ByteOrder::native();

tests/decode_images.rs

@@ -1,11 +1,11 @@
 extern crate tiff;
 
 use tiff::decoder::{ifd, Decoder, DecodingSampleBuffer};
-use tiff::tags::{ByteOrder, Type};
+use tiff::tags::{ByteOrder, PhotometricInterpretation, Type};
 use tiff::ColorType;
 
 use std::fs::File;
-use std::io::{Read, Seek};
+use std::io::{Cursor, Read, Seek};
 use std::path::PathBuf;
 
 const TEST_IMAGE_DIR: &str = "./tests/images/";
@@ -1059,3 +1059,126 @@ fn test_gray_u12_hpredict() {
     // Delta-encoded 12-bit values: [100, 200, 300, 400, 1000, 2000, 3000, 4000]
     test_image_sum_u16("hpredict-1c-12b.tiff", ColorType::Gray(12), 11000);
 }
+
+// ---- TransparencyMask ----
+
+/// Build a minimal uncompressed TIFF in memory with the given photometric interpretation.
+/// Returns a 2x2 image with the provided pixel data as a single strip.
+fn build_raw_tiff(
+    photometric: PhotometricInterpretation,
+    bits_per_sample: u16,
+    samples_per_pixel: u16,
+    pixel_data: &[u8],
+) -> Vec<u8> {
+    let width: u16 = 2;
+    let height: u16 = 2;
+    let num_tags: u16 = 8;
+    let ifd_offset: u32 = 8;
+    let ifd_size = 2 + (num_tags as u32) * 12 + 4;
+    let strip_offset = 8 + ifd_size;
+
+    let mut buf = Vec::new();
+    buf.extend_from_slice(b"II");
+    buf.extend_from_slice(&42u16.to_le_bytes());
+    buf.extend_from_slice(&ifd_offset.to_le_bytes());
+    buf.extend_from_slice(&num_tags.to_le_bytes());
+
+    // IFD entries (must be ascending tag order). Tag IDs stay raw for the
+    // ascending-order property; the field type uses the `Type` enum.
+    for &(tag, typ, value) in &[
+        (256u16, Type::SHORT, width as u32),             // ImageWidth
+        (257, Type::SHORT, height as u32),               // ImageLength
+        (258, Type::SHORT, bits_per_sample as u32),      // BitsPerSample
+        (259, Type::SHORT, 1),                           // Compression = None
+        (262, Type::SHORT, photometric.to_u16() as u32), // PhotometricInterpretation
+        (273, Type::LONG, strip_offset),                 // StripOffsets
+        (277, Type::SHORT, samples_per_pixel as u32),    // SamplesPerPixel
+        (279, Type::LONG, pixel_data.len() as u32),      // StripByteCounts
+    ] {
+        buf.extend_from_slice(&tag.to_le_bytes());
+        buf.extend_from_slice(&typ.to_u16().to_le_bytes());
+        buf.extend_from_slice(&1u32.to_le_bytes());
+        buf.extend_from_slice(&value.to_le_bytes());
+    }
+
+    buf.extend_from_slice(&0u32.to_le_bytes()); // next IFD = 0
+    buf.extend_from_slice(pixel_data);
+    buf
+}
+
+#[test]
+fn test_transparency_mask() {
+    // 2x2 mask: values 0, 255, 128, 64
+    let data = build_raw_tiff(
+        PhotometricInterpretation::TransparencyMask,
+        8,
+        1,
+        &[0, 255, 128, 64],
+    );
+    let mut decoder = Decoder::open(Cursor::new(&data)).unwrap();
+    decoder.next_image().unwrap();
+
+    assert_eq!(decoder.colortype().unwrap(), ColorType::Gray(8));
+
+    let result = decoder.read_image().unwrap();
+    match result {
+        DecodingSampleBuffer::U8(pixels) => {
+            // TransparencyMask is decoded as grayscale
+            assert_eq!(pixels.len(), 4);
+            assert_eq!(pixels[0], 0);
+            assert_eq!(pixels[1], 255);
+            assert_eq!(pixels[2], 128);
+            assert_eq!(pixels[3], 64);
+        }
+        _ => panic!("Expected U8 buffer"),
+    }
+}
+
+// ---- CIE L*a*b* ----
+
+#[test]
+fn test_cielab_decode() {
+    // 2x2 CIELab image: 3 bytes per pixel (L, a, b)
+    // L=128, a=0 (signed neutral), b=0 (signed neutral)
+    // After re-biasing: a=128 (unsigned neutral), b=128 (unsigned neutral)
+    let pixels: Vec<u8> = vec![
+        128, 0, 0, // pixel 0: L=128, a=0, b=0 (neutral gray)
+        255, 50, 50, // pixel 1: L=255, a=+50, b=+50
+        100, 206, 206, // pixel 2: L=100, a=-50 (as i8), b=-50 (as i8)
+        0, 128, 128, // pixel 3: L=0, a=-128 (as i8), b=-128 (as i8)
+    ];
+
+    let data = build_raw_tiff(PhotometricInterpretation::CIELab, 8, 3, &pixels);
+    let mut decoder = Decoder::open(Cursor::new(&data)).unwrap();
+    decoder.next_image().unwrap();
+
+    assert_eq!(decoder.colortype().unwrap(), ColorType::Lab(8));
+
+    let result = decoder.read_image().unwrap();
+    match result {
+        DecodingSampleBuffer::U8(decoded) => {
+            assert_eq!(decoded.len(), 12); // 4 pixels * 3 channels
+
+            // Pixel 0: L=128, a=0+128=128, b=0+128=128
+            assert_eq!(decoded[0], 128);
+            assert_eq!(decoded[1], 128);
+            assert_eq!(decoded[2], 128);
+
+            // Pixel 1: L=255, a=50+128=178, b=50+128=178
+            assert_eq!(decoded[3], 255);
+            assert_eq!(decoded[4], 178);
+            assert_eq!(decoded[5], 178);
+
+            // Pixel 2: L=100, a=206+128=78 (wrapping), b=206+128=78
+            assert_eq!(decoded[6], 100);
+            assert_eq!(decoded[7], 78);
+            assert_eq!(decoded[8], 78);
+
+            // Pixel 3: L=0, a=128+128=0 (wrapping), b=128+128=0
+            assert_eq!(decoded[9], 0);
+            assert_eq!(decoded[10], 0);
+            assert_eq!(decoded[11], 0);
+        }
+        _ => panic!("Expected U8 buffer"),
+    }
+}