Added some experimental data about Huffman code lengths.

diff --git a/codelen.txt b/codelen.txt
new file mode 100644 (file)
index 0000000..279ffdc
--- /dev/null
+++ b/codelen.txt
@@ -0,0 +1,35 @@
+Huffman code lengths for a typical YCbCr JPEG (photography):
+
+0: 0 (0.00%)
+1: 0 (0.00%)
+2: 7664341 (59.62%)
+3: 1447176 (70.87%)
+4: 1510109 (82.62%)
+5: 737949 (88.36%)
+6: 759048 (94.26%)
+7: 426433 (97.58%)
+8: 156391 (98.80%)
+9: 88996 (99.49%)
+10: 30697 (99.73%)
+11: 9959 (99.81%)
+12: 12054 (99.90%)
+13: 5962 (99.95%)
+14: 1574 (99.96%)
+15: 0 (99.96%)
+16: 5377 (100.00%)
+17: 0 (100.00%)
+18: 0 (100.00%)
+19: 0 (100.00%)
+20: 0 (100.00%)
+21: 0 (100.00%)
+22: 0 (100.00%)
+23: 0 (100.00%)
+24: 0 (100.00%)
+25: 0 (100.00%)
+26: 0 (100.00%)
+27: 0 (100.00%)
+28: 0 (100.00%)
+29: 0 (100.00%)
+30: 0 (100.00%)
+31: 0 (100.00%)
+