Tesseract Provider

Prerequisites

You must have tesseract installed along with language packs. For example in Arch Linux

yay tesseract

3 extra/tesseract-data-ben 2:4.1.0-4 (1.3 MiB 1.7 MiB) [tesseract-data]
    Tesseract OCR data (ben)
2 extra/tesseract-data-jpn 2:4.1.0-4 (14.1 MiB 34.0 MiB) [tesseract-data]
    Tesseract OCR data (jpn)
1 extra/tesseract 5.5.1-1 (1.6 MiB 4.6 MiB) (Installed) <-----------
    An OCR program

# language packs
102 extra/tesseract-data-eng 2:4.1.0-4 (9.0 MiB 22.4 MiB) [tesseract-data] (Installed) # english
35 extra/tesseract-data-ell 2:4.1.0-4 (3.0 MiB 7.2 MiB) [tesseract-data] (Installed) # greek

Most Basic Schema

~/.config/gowall/schema.yml

schemas:
  - name: "tes"
    config:
      ocr:
        provider: "tesseract"
        model: "tesseract"

gowall ocr img.png -s tes

Please refer to the schemas section for more information on how to use schemas with rate limiting,pre/post processing options,text correction etc...

Tesseract + LLM Grammar/Format Correction

You can checkout the text correction example in the schema section where it showcases how tesseract does the OCR and the LLM corrects the grammar mistakes, closes brackets etc...

Example

Take the below image for example, lets compare the plain tesseract output and the tesseract + LLM corrected output of the extracted text.

Specimen

gowall ocr img.png -p tesseract -m tesseract 

Plain Tesseract output

// input

nums := [lint{2, 3, 4, 7, 1}

// stage 1

dataChannel := sliceToChannel(nums)

// stage 2

finalChannel := sq(dataChannel)

// stage 3

for n := range finalChannel {
fmt.Println(n)

As we can see Tesseract did several mistakes :

1. instead of []int it used [lint, hell it didn't even bother to close the brackets.
2. There is a missing closing bracket } at the end in the for range loop.
3. the for n:= and the fmt.Println() are in the same vertical line, which isn't correct since the println should be one tab in.

Lets try the tesseract + LLM corrected output, to see what exactly changes now that tesseracts output is given to a model.

gowall ocr img.png -s tes-llm

Tesseract + LLM corrected output

```go
// input

nums := []int{2, 3, 4, 7, 1}

// stage 1

dataChannel := sliceToChannel(nums)

// stage 2

finalChannel := sq(dataChannel)

// stage 3

for n := range finalChannel {
    fmt.Println(n)
}
```

Seems that things are a lot better now.

1. The model knows go so it infered from the syntax and context that it should be []int not [lint.

2. Actually closed the brackets } at the end in the for range loop.

3. the for n:= and the fmt.Println() are now in the correct vertical line.

4. It even threw a little bonus wrapping the code inside a markdown code block and adding a language tag.

The above was done with the following schema :

~/.config/gowall/schema.yml

  - name: "tes-llm"
    config:
      ocr:
        provider: "tesseract"
        model: "tesseract"
      text_correction:
        enabled: true
        rate_limit:  # <-- rate limits the post-processing provider not the ocr provider
            rps: 2
            burst: 4
        provider:
          provider: "openrouter"  # You can specify all the provider configs as seen in the 'Provider Configs' section
          model: "x-ai/grok-4-fast:free"

Obviously here is where a selfhosted llm could come to play for grammar correction, since you don't need beefy specs to run a simple model for this task. The example above is just for demonstration purposess.