otf-pixels-codec-jpeg 0.2.0

Baseline JPEG codec for otf-pixels, implemented from scratch.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214

//! The example tables from ITU-T T.81 Annex K, which encoders use as defaults.
//!
//! The standard calls these "example" tables and requires no encoder to use
//! them. In practice essentially every baseline encoder does, because they
//! were derived from perceptual experiments and because a decoder that has
//! seen them a billion times has nothing to gain from a bespoke set.
//!
//! Deriving optimal Huffman tables from the actual coefficient statistics
//! would save a few percent, at the cost of buffering the whole image to
//! count symbols before the first byte can be written. That trade runs
//! directly against ADR-0005's streaming contract, so it is not made here.

/// Luminance quantization steps, in natural (row-major) order — Table K.1.
///
/// The values rise towards the bottom right because that is where the high
/// frequencies live, and the eye does not miss them.
pub const LUMA_QUANT: [u16; 64] = [
    16, 11, 10, 16, 24, 40, 51, 61, //
    12, 12, 14, 19, 26, 58, 60, 55, //
    14, 13, 16, 24, 40, 57, 69, 56, //
    14, 17, 22, 29, 51, 87, 80, 62, //
    18, 22, 37, 56, 68, 109, 103, 77, //
    24, 35, 55, 64, 81, 104, 113, 92, //
    49, 64, 78, 87, 103, 121, 120, 101, //
    72, 92, 95, 98, 112, 100, 103, 99,
];

/// Chrominance quantization steps, in natural order — Table K.2.
///
/// Flat at 99 across most of the block: colour detail is discarded far more
/// aggressively than brightness detail, which is the same perceptual fact
/// that makes chroma subsampling acceptable.
pub const CHROMA_QUANT: [u16; 64] = [
    17, 18, 24, 47, 99, 99, 99, 99, //
    18, 21, 26, 66, 99, 99, 99, 99, //
    24, 26, 56, 99, 99, 99, 99, 99, //
    47, 66, 99, 99, 99, 99, 99, 99, //
    99, 99, 99, 99, 99, 99, 99, 99, //
    99, 99, 99, 99, 99, 99, 99, 99, //
    99, 99, 99, 99, 99, 99, 99, 99, //
    99, 99, 99, 99, 99, 99, 99, 99,
];

/// Luminance DC code lengths and symbols — Table K.3.
pub const LUMA_DC_COUNTS: [u8; 16] = [0, 1, 5, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0];
/// Symbols for [`LUMA_DC_COUNTS`]: the twelve magnitude categories.
pub const LUMA_DC_VALUES: [u8; 12] = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11];

/// Chrominance DC code lengths and symbols — Table K.4.
pub const CHROMA_DC_COUNTS: [u8; 16] = [0, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0];
/// Symbols for [`CHROMA_DC_COUNTS`].
pub const CHROMA_DC_VALUES: [u8; 12] = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11];

/// Luminance AC code lengths — Table K.5.
pub const LUMA_AC_COUNTS: [u8; 16] = [0, 2, 1, 3, 3, 2, 4, 3, 5, 5, 4, 4, 0, 0, 1, 0x7D];

/// Luminance AC symbols, each a run of zeros in the high nibble and a
/// magnitude category in the low one — Table K.5.
pub const LUMA_AC_VALUES: [u8; 162] = [
    0x01, 0x02, 0x03, 0x00, 0x04, 0x11, 0x05, 0x12, //
    0x21, 0x31, 0x41, 0x06, 0x13, 0x51, 0x61, 0x07, //
    0x22, 0x71, 0x14, 0x32, 0x81, 0x91, 0xA1, 0x08, //
    0x23, 0x42, 0xB1, 0xC1, 0x15, 0x52, 0xD1, 0xF0, //
    0x24, 0x33, 0x62, 0x72, 0x82, 0x09, 0x0A, 0x16, //
    0x17, 0x18, 0x19, 0x1A, 0x25, 0x26, 0x27, 0x28, //
    0x29, 0x2A, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, //
    0x3A, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, //
    0x4A, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, //
    0x5A, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, //
    0x6A, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, //
    0x7A, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, //
    0x8A, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, //
    0x99, 0x9A, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7, //
    0xA8, 0xA9, 0xAA, 0xB2, 0xB3, 0xB4, 0xB5, 0xB6, //
    0xB7, 0xB8, 0xB9, 0xBA, 0xC2, 0xC3, 0xC4, 0xC5, //
    0xC6, 0xC7, 0xC8, 0xC9, 0xCA, 0xD2, 0xD3, 0xD4, //
    0xD5, 0xD6, 0xD7, 0xD8, 0xD9, 0xDA, 0xE1, 0xE2, //
    0xE3, 0xE4, 0xE5, 0xE6, 0xE7, 0xE8, 0xE9, 0xEA, //
    0xF1, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7, 0xF8, //
    0xF9, 0xFA,
];

/// Chrominance AC code lengths — Table K.6.
pub const CHROMA_AC_COUNTS: [u8; 16] = [0, 2, 1, 2, 4, 4, 3, 4, 7, 5, 4, 4, 0, 1, 2, 0x77];

/// Chrominance AC symbols — Table K.6.
pub const CHROMA_AC_VALUES: [u8; 162] = [
    0x00, 0x01, 0x02, 0x03, 0x11, 0x04, 0x05, 0x21, //
    0x31, 0x06, 0x12, 0x41, 0x51, 0x07, 0x61, 0x71, //
    0x13, 0x22, 0x32, 0x81, 0x08, 0x14, 0x42, 0x91, //
    0xA1, 0xB1, 0xC1, 0x09, 0x23, 0x33, 0x52, 0xF0, //
    0x15, 0x62, 0x72, 0xD1, 0x0A, 0x16, 0x24, 0x34, //
    0xE1, 0x25, 0xF1, 0x17, 0x18, 0x19, 0x1A, 0x26, //
    0x27, 0x28, 0x29, 0x2A, 0x35, 0x36, 0x37, 0x38, //
    0x39, 0x3A, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, //
    0x49, 0x4A, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, //
    0x59, 0x5A, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, //
    0x69, 0x6A, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, //
    0x79, 0x7A, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, //
    0x88, 0x89, 0x8A, 0x92, 0x93, 0x94, 0x95, 0x96, //
    0x97, 0x98, 0x99, 0x9A, 0xA2, 0xA3, 0xA4, 0xA5, //
    0xA6, 0xA7, 0xA8, 0xA9, 0xAA, 0xB2, 0xB3, 0xB4, //
    0xB5, 0xB6, 0xB7, 0xB8, 0xB9, 0xBA, 0xC2, 0xC3, //
    0xC4, 0xC5, 0xC6, 0xC7, 0xC8, 0xC9, 0xCA, 0xD2, //
    0xD3, 0xD4, 0xD5, 0xD6, 0xD7, 0xD8, 0xD9, 0xDA, //
    0xE2, 0xE3, 0xE4, 0xE5, 0xE6, 0xE7, 0xE8, 0xE9, //
    0xEA, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7, 0xF8, //
    0xF9, 0xFA,
];

/// Scale a quantization table for a quality setting, 1..=100.
///
/// The mapping is the IJG one, and is worth keeping bit-compatible: quality
/// numbers are compared across tools constantly, and "quality 80" meaning
/// something different here than in every other encoder would be a trap.
/// Quality 50 leaves the tables untouched; below it they scale up steeply,
/// above it they flatten towards 1.
#[must_use]
pub fn scale_quant(base: &[u16; 64], quality: u8) -> [u16; 64] {
    let quality = u32::from(quality.clamp(1, 100));
    let scale = if quality < 50 {
        5000 / quality
    } else {
        200 - quality * 2
    };
    std::array::from_fn(|index| {
        let value = u32::from(*base.get(index).unwrap_or(&1));
        // Steps are clamped to 1..=255 so the table fits an 8-bit DQT, and
        // so a step of zero — which would divide by zero — cannot arise.
        ((value * scale + 50) / 100).clamp(1, 255) as u16
    })
}

#[cfg(test)]
#[allow(
    clippy::unwrap_used,
    clippy::indexing_slicing,
    reason = "tests operate on known-good values and assert shapes directly"
)]
mod tests {
    use super::*;
    use crate::huffman::HuffmanTable;

    #[test]
    fn the_standard_tables_are_valid_prefix_codes() {
        // Building a *decoding* table from each set proves the counts and
        // symbol lists agree and that no set is over-subscribed. A
        // transcription slip in 162 hex values would otherwise sit here
        // silently until a decoder somewhere rejected our output.
        for (counts, values) in [
            (&LUMA_DC_COUNTS, &LUMA_DC_VALUES[..]),
            (&CHROMA_DC_COUNTS, &CHROMA_DC_VALUES[..]),
            (&LUMA_AC_COUNTS, &LUMA_AC_VALUES[..]),
            (&CHROMA_AC_COUNTS, &CHROMA_AC_VALUES[..]),
        ] {
            let total: usize = counts.iter().map(|&c| c as usize).sum();
            assert_eq!(total, values.len(), "counts and symbols disagree");
            HuffmanTable::new(counts, values.to_vec()).unwrap();
        }
    }

    #[test]
    fn ac_symbol_sets_are_complete_and_unique() {
        // Every (run, size) pair an encoder can emit must have a code. The
        // legal set is size 1..=10 for runs 0..=15, plus ZRL and EOB.
        for values in [&LUMA_AC_VALUES[..], &CHROMA_AC_VALUES[..]] {
            let mut seen = std::collections::HashSet::new();
            for &symbol in values {
                assert!(seen.insert(symbol), "duplicate AC symbol {symbol:#04x}");
            }
            assert!(seen.contains(&0x00), "no end-of-block code");
            assert!(seen.contains(&0xF0), "no zero-run-length code");
            for run in 0..16_u8 {
                for size in 1..=10_u8 {
                    let symbol = (run << 4) | size;
                    assert!(seen.contains(&symbol), "no code for {symbol:#04x}");
                }
            }
        }
    }

    #[test]
    fn quality_fifty_leaves_the_base_tables_alone() {
        // The definition of the IJG scale: 50 is the identity point.
        assert_eq!(scale_quant(&LUMA_QUANT, 50), LUMA_QUANT);
        assert_eq!(scale_quant(&CHROMA_QUANT, 50), CHROMA_QUANT);
    }

    #[test]
    fn quality_moves_the_steps_the_right_way() {
        let low = scale_quant(&LUMA_QUANT, 10);
        let high = scale_quant(&LUMA_QUANT, 95);
        // Coarser steps at low quality, finer at high, everywhere.
        for index in 0..64 {
            assert!(low[index] >= LUMA_QUANT[index].min(255), "index {index}");
            assert!(high[index] <= LUMA_QUANT[index], "index {index}");
        }
        // Quality 100 is the finest the mapping allows: every step is 1.
        assert_eq!(scale_quant(&LUMA_QUANT, 100), [1_u16; 64]);
    }

    #[test]
    fn every_step_stays_in_the_eight_bit_dqt_range() {
        // A step of zero would divide by zero; a step above 255 would not fit
        // the 8-bit table we write.
        for quality in 1..=100_u8 {
            for base in [&LUMA_QUANT, &CHROMA_QUANT] {
                for step in scale_quant(base, quality) {
                    assert!((1..=255).contains(&step), "quality {quality}: {step}");
                }
            }
        }
    }
}