'confidence' for single character is different between 'MeanTextConf()` and 'Confidence(level)'

[NewUserHa]

Current Behavior

It's unconvinent to compile exe so I here used c++ bind from python. (I examed its code and it's directly using TessBaseAPI in C++)

api = tesserocr.PyTessBaseAPI(r"", 'chi_sim', 10)
api.SetVariable("save_blob_choices", "T")
api.SetVariable("lstm_choice_mode", "2")

api.SetImageFile(r"...")
api.Recognize()

choices = []
ri = api.GetIterator()
level = tesserocr.RIL.SYMBOL
for r in tesserocr.iterate_level(ri, level):
    choices.append([r.GetUTF8Text(level), r.Confidence(level)])
    for _ in r.GetChoiceIterator():
        choices.append([_.GetUTF8Text(), _.Confidence()])

print([(_.GetUTF8Text(), _.Confidence()) for _ in api.GetIterator().GetChoiceIterator()])
print(api.GetUTF8Text().strip(), api.MeanTextConf(), choices, api.MapWordConfidences() if api.GetUTF8Text() else 0)

api.End()

output:

[('员', 91.3373794555664), ('灵', 0.0)]
员 93 [['员', 99.01185607910156], ['员', 91.3373794555664], ['灵', 0.0]] [('员', 93)]

[('会', 92.27279663085938), ('针', 0.0), ('，', 0.0), ('|', 0.0), ('。', 0.0)]
会 96 [['会', 99.53471374511719], ['会', 92.27279663085938], ['针', 0.0], ['，', 0.0], ['|', 0.0], ['。', 0.0]] [('会', 96)]

[('患', 64.94859313964844), ('理', 0.0), ('写', 0.0), ('上', 0.0), ('雪', 0.0), ('由', 0.0)]
患 0 [['患', 80.64909362792969], ['患', 64.94859313964844], ['理', 0.0], ['写', 0.0], ['上', 0.0], ['雪', 0.0], ['由', 0.0]] [('患', 0)]

for most of the single characters, the MeanTextConf() returned seems to be the mean of (like '会' (99 + 92)/2 ~= 96), but for the first case '员' (99+91)/2 ~= 95, rather than 93 as the shown above.
moreover, the single character "患" has both MapWordConfidences and MeanTextConf 0 (zero), and obviously that's wrong.

pictures of above examples: (all are extracted from font files with same point size)

Expected Behavior

in the above example, since it's a single character, the confidence should be '99.01185607910156' '99.53471374511719' (those returned by r.Confidence(level)), because that's the correct character.

Suggested Fix

as above

tesseract -v

bunded, but:
tesseract 5.3.1
leptonica-1.83.1 (Jun 13 2023, 19:19:21) [MSC v.1935 LIB Release x64]
libgif 5.2.1 : libjpeg 6b (libjpeg-turbo 2.1.4) : libpng 1.6.39 : libtiff 4.5.0 : zlib 1.2.13 : libwebp 1.3.0 : libopenjp2 2.5.0

Operating System

Windows 10

Other Operating System

No response

uname -a

No response

Compiler

No response

CPU

No response

Virtualization / Containers

No response

Other Information

No response

[zdenop]

We do not support 3rd party projects (tesseract wrappers).
Please reproduce the problem with C++ code (full working code + input image for testing and maybe desired output, AFAIK we do not active Chinese developer).
Also, provide information if some previous version (e.g. 4.x) was working as you expected.
Thanks.