CameraX ImageAnalysis YUV‑to‑RGB Conversion Using libyuv and NDK

Background CameraX is a Jetpack library that simplifies camera app development on Android. It can work with ML Kit or TensorFlow Lite for tasks such as text recognition and image labeling, but converting image data from YUV to RGB has historically been cumbersome.

The new feature adds direct YUV‑420_888 to RGBA_8888 conversion in the ImageAnalysis pipeline, allowing developers to choose either YUV or RGB output for downstream processing.

YUV_420_888 Format This format provides three separate planes (Y, U, V) with 8‑bit luma and chroma components. It supports many OEM variations and is widely used in ImageAnalysis, but many applications prefer the simpler RGB format.

RGBA_8888 Format RGBA_8888 is a standard 4‑channel format (red, green, blue, alpha) with 8 bits per channel, making it easy to feed into TensorFlow Lite models.

API Implementation Three conversion approaches were evaluated:

Java/Kotlin implementation – simple but slow and creates garbage‑collection pressure.

RenderScript – optimized for compute‑intensive tasks but deprecated starting with Android 12.

Native solution using libyuv and the NDK – chosen for performance and future compatibility.

The native pipeline uses libyuv for scaling, rotation, and YUV‑to‑RGB conversion, then writes the result to an ImageProxy via ImageReader / ImageWriter and ANativeWindow , preserving compatibility with older API levels.

Output Formats Starting with CameraX 1.1.0‑alpha08, developers can set the output format in ImageAnalysis via setOutputImageFormat to either YUV_420_888 or RGBA_8888. RGBA_8888 uses PixelFormat.RGBA_8888 , while YUV_420_888 uses ImageFormat.YUV_420_888 .

Performance Benchmarks on various devices and Android versions show that the libyuv‑based pipeline consistently outperforms the RenderScript implementation across different resolutions.

Conclusion CameraX now supports YUV‑to‑RGB conversion directly in the ImageAnalysis pipeline, giving developers a simple way to select the desired output format and integrate with ML Kit, TensorFlow Lite, or custom native models. Future work will add more image‑processing capabilities to other use cases such as ImageCapture and Preview.

References

CameraX documentation: https://developer.android.google.cn/training/camerax

ML Kit: https://developers.google.cn/ml-kit

TensorFlow Lite: https://www.tensorflow.org/lite

libyuv source: https://chromium.googlesource.com/libyuv/libyuv/

CameraX release notes: https://developer.android.google.cn/jetpack/androidx/releases/camera