FPix

FPix is a floating-point image representation used in Leptonica, a C library for image processing. An FPix stores a two-dimensional grayscale image where each pixel value is a 32-bit floating point number. This format is used to hold intermediate results of image processing that require fractional precision, such as convolution outputs, gradient magnitudes, normalized intensities, or other computations that exceed the 8-bit range of standard pixel images.

Internally an FPix contains width and height, and a contiguous array of float values laid out row-major. Access is by (x, y) coordinates; the pixel value is typically addressed as data[y * width + x]. FPix is designed to work alongside PIX (the 8-bit/16-bit image type in Leptonica) and routines exist to convert between FPix and PIX, as well as to read/write FPix data to files or memory. Creation routines allocate the necessary memory, and a destructor frees it.

Common uses include storing pre- and post-processing results in OCR pipelines, color and intensity normalization, and filter responses prior to binarization or feature extraction. Because FPix holds floating point data, it is well-suited for high-precision analysis, but it consumes more memory than integer-based images. Limitations include higher memory usage and the need for careful handling of scaling, since subsequent stages may expect normalized values. FPix is one of several image representations in Leptonica and is often used internally by algorithms requiring floating point precision.