Struct OcrEngine 

pub struct OcrEngine { /* private fields */ }

Detects and recognizes text in images.

OcrEngine uses machine learning models to detect text, analyze layout and recognize text in an image.

Implementations

impl OcrEngine

pub fn new(params: OcrEngineParams) -> Result<OcrEngine>

Construct a new engine from a given configuration.

pub fn prepare_input(&self, image: ImageSource<'_>) -> Result<OcrInput>

Preprocess an image for use with other methods of the engine.

pub fn detect_words(&self, input: &OcrInput) -> Result<Vec<RotatedRect>>

Detect text words in an image.

Returns an unordered list of the oriented bounding rectangles of each word found.

pub fn detect_text_pixels(&self, input: &OcrInput) -> Result<NdTensor<f32, 2>>

Detect text pixels in an image.

Returns an (H, W) tensor indicating the probability of each pixel in the input being part of a text word. This is a low-level API that is useful for debugging purposes. Use detect_words for a higher-level API that returns oriented bounding boxes of words.

pub fn find_text_lines( &self, _input: &OcrInput, words: &[RotatedRect], ) -> Vec<Vec<RotatedRect>>

Perform layout analysis to group words into lines and sort them in reading order.

words is an unordered list of text word rectangles found by OcrEngine::detect_words. The result is a list of lines, in reading order. Each line is a sequence of word bounding rectangles, in reading order.

pub fn recognize_text( &self, input: &OcrInput, lines: &[Vec<RotatedRect>], ) -> Result<Vec<Option<TextLine>>>

Recognize lines of text in an image.

lines is an ordered list of the text line boxes in an image, produced by OcrEngine::find_text_lines.

The output is a list of TextLines corresponding to the input image regions. Entries can be None if no text was found in a given line.

pub fn prepare_recognition_input( &self, input: &OcrInput, line: &[RotatedRect], ) -> Result<NdTensor<f32, 2>>

Prepare an image for input into the text line recognition model.

This method exists to help with debugging recognition issues by exposing the preprocessing that OcrEngine::recognize_text does before it feeds an image into the recognition model. Use OcrEngine::recognize_text to recognize text.

line is a sequence of RotatedRects that make up a line of text.

Returns a greyscale (H, W) image with values in [-0.5, 0.5].

pub fn detection_threshold(&self) -> f32

Return the confidence threshold applied to the output of the text detection model to determine whether a pixel is text or not.

pub fn get_text(&self, input: &OcrInput) -> Result<String>

Convenience API that extracts all text from an image as a single string.