QHdrCamera component for High Dynamic Range Imaging
This article explains how to use QHdrCamera, a custom QML component for creating High Dynamic Range Images using the device camera.
Note: This is an entry in the PureView Imaging Competition 2012Q2
High Dynamic Range Imaging (HDRI or HDR) is a set of methods used in image processing, computer graphics, and photography, to allow a greater dynamic range between the lightest and darkest areas of an image.
Non-HDR cameras take pictures at one exposure level with a limited contrast range. This results in the loss of detail in bright or dark areas of a picture, depending on whether the camera had a low or high exposure setting. HDR compensates for this loss of detail by taking multiple pictures at different exposure levels and intelligently stitching them together so that we eventually arrive at a picture that is representative in both dark and bright areas.
This article presents custom QML components that can be used to capture and preview an HDR image. It includes demonstration images taken using the components, an explanation of how to use the component in your projects, and an overview of the algorithm used.
Images taken with a Nokia N8
The same images taken with an iPhone 4Gs using his built in HDR technology
Images taken with a Nokia N8 with very bad light conditions:
Performance & Limitations
Taking the three "seed" images takes significant time, as does the subsequent calculation of the final HDR image. The table below shows the approximate time (in seconds) to take the initial photos and then to create an HDR image on the Nokia N8.
|Resolution||Shooting time||Processing time|
All operations are time consuming and require a huge amount of resources:
- CPU. The 600Mhz processor on the Nokia N8 is not particularly fast, so we usually have to work with low resolution images in order to keep the calculation time down to an acceptable level. The1.3GHz processor on the Nokia 808 should allow higher resolution images to be calculated in a reasonable time.
- Memory. Starting from 256MB of available RAM it's impossible to allow very high resolutions.
- Mobile devices can't provide the same exposure range compared to professional cameras.
Despite the above limitations, the HDR element demonstrated by this article can provide much better images in difficult light conditions than that provided by the default camera app. After all, much better to capture those special moments at some resolution, than to get a blank image at higher resolution.
Recommendations for using HDR
HDR is recommended for the following use-cases:
- Landscape: A classic use of HDR in regular photography is on landscape shots with a bright sky above the horizon line and darker foreground below.
- Outdoor portraits: The midday sun’s harsh light usually makes for unflattering portraits. It can cast strong shadows on a person’s face and create dark circles around the eyes. It also bounces off of skin and accentuates shiny spots. HDR can minimize the effect of these two extremes and create a more evenly lit portrait.
HDR is less effective for:
- Capturing motion: To create an HDR image, QHdrCamera takes three photos in quick' succession. If you're taking a photo of a fast moving subject or if you move the phone while shooting, the final HDR image will show "ghosting". That's when the multiple images aren't aligned and objects appear in multiple places. If you experience this problem frequently with your HDR images, consider mounting the phone on a tripod.
- Capturing vivid colors : The HDR mode can bring colors back into blown-out or dark areas. But when photographing brightly colored subjects that are properly exposed, HDR results in a disappointing desaturation. If the allure of your image is that it shows vivid colors, HDR is not the best choice.
Using the component
- Download source code from QHdrCamera
- Include QHdrCamera directory into your project
symbian: TARGET.EPOCHEAPSIZE = 0x20000 0xC000000
We highly recommend increasing the heap size available to the app, as shown above.
This code shows how to add the HdrCamera Element so that it is available for use within QML.
Q_DECL_EXPORT int main(int argc, char *argv)
QScopedPointer<QApplication> app(createApplication(argc, argv));
qmlRegisterType<QHdrCamera>("com.imaging.hdr", 1, 0, "QHdrCamera");
// Resize the root QML element to view size
// Performance optimization flags
The following code shows how to use the QHdrCamera QML element:
- Declare the component along with a resolution and exposure range if the default values (shown commented out below) are not acceptable.
- Call captureImage() to start taking the images.
- Component provide a built in preview management through boolean preview property, anyway you can implement onFinished() to display or handle the image once all calculation has been finished. In this example we've used the custom ImagePreview component to display the image, based on the onFinished() signal's parameters pathResultImage and pathOriginalImage. There is nothing to stop you rolling out your own viewer component.
- availableDevices is an array property with the list of all available camera devices on your phone. Typically on first entry you will find the main camera.
import QtQuick 1.1
import com.nokia.symbian 1.1
//import com.nokia.meego 1.0
import com.imaging.hdr 1.0
busy.running = true; busy.visible = true;
busy.running = false; busy.visible = false;
camera.preview = true;
previewb.text = "Close Preview"
text: "Show Preview"
camera.preview = !camera.preview
if( camera.preview )
previewb.text = "Close Preview"
previewb.text = "Show Preview"
QHdrCamera takes three photos in a sequence when captureImage() is called, each with a different exposure level based on the range specified in the exposureRange property.
The resulting HDR image is saved on default's phone image directory with the prefix hdr, followed by a timestamp, in a jpeg format.
Signal finished(string pathResultImage, string pathOriginalImage) provides information on where image is saved so that it can be displayed in the preview.
ImagePreview QML element provides a simple HDR image preview: users swipe left or right cycle between the original (exposure=0) and resulting HDR image. This is implemented as a ListView.
- exposureRange : real - sets the exposure working range. The default value of 2 means that three photos will be taken: the first at exposure 0, the second with exposure +2 and the third with exposure -2. Setting 1 the range will be -1,0,1.
- captureResolution : size - The resolution to capture the image at. If empty, the system will pick a resolution of 800x600
- captureImage() - starts the process of capturing three images with different exposure ranges.
- onFinished(string pathResultImage, string pathOriginalImage) - handler for the finished(string pathResultImage, string pathOriginalImage) signal. Invoked when the rendering process is finished.
Variable pathResultImage contains the location of the resulted saved image, pathOriginalImage contains the location of the original saved image, useful for the preview process.
HDR Algorithm Overview
The algorithm I've used comes from the paper "Multi-exposure Imaging on Mobile Devices", a collaboration between Nokia Research Center and Stanford University. The reference provides full details - but here is a brief overview.
To provide an HDR Image we need to record a number of images at different exposures, and then merge them using an exposure fusion algorithm. Exposure fusion computes a scalar-valued weight map for each image, and performs a weighted blend of inputs to obtain the final result.
The blending is computed using a multi-resolution algorithm that build a Laplacian pyramid of the inputs and a Gaussian pyramid of the maps, blends each layer separately and then collapes the resulting Laplacian pyramid to obtain the result.
Let ( I1 .. In ) the stack of n input images captured by the camera ( in that context just three ). The weight for each pixel (i,j) of image k is computed as:
Current parameters used are μ=0.5 and σ=0.2.
Tone Mapping. The result of Gauss filter (1) for each image. Interesting to note the different light details on each image.
Laplace application on Gauss filter result:
The result consists of multiplyng the images by their weight maps and blending:
Images taken with a Nokia N8 using QHdrCamera QML Element
- process() - implements all steps described above
- gauss() - implements all steps described by formula (1)
- laplace() - implements the Laplacian operation
This article describes the first "alpha" version of the project. While a huge amount of work has been done to make it an effective "proof of concept" there is still a lot more that could be done. For example:
- Dynamically exposure range calculation
- Better camera motion compensation algorithm
- Inclusion of metadata information inside the header image like gps location
- Generic performance improvements
I plan to continue this development based on user feedback and interest. Please add your comments in the Comments section below.
- ↑ Multi-exposure Imaging on Mobile Devices Natasha Gelfand, Andrew Adams, Sung Hee Park, Kari Pulli. 2010