- To achieve seamless integration of the real scene with virtual objects, we need to obtain accurate illumination condition of the real scene. Using HDRI (High Dynamic Range Image) we can acquire and control real world illumination information. There are various intensity levels of light sources in a HDRI; some regions are very bright and others are not. Bright regions contain much more lighting information than the other regions. Thus in this paper we propose a light source estimation method by using the ratio of intensity of radiation from segmented HDR images which are divided by characteristics of histogram value. This paper describes three mains steps performed in the research. First, we segment an image efficiently; second, we analyze the distribution of intensity of radiation. Finally, we estimate light sources from segmented HDR images by using the ratio of intensity of radiation.
Icommercial products, game, VR and AR, seamless n various fields such as a movie, advertisement for integration of virtual objects with a real world scene has been frequently used. To achieve more realistic synthetic results, accurate geometric models, material appearance and real world illumination conditions are required. Among them, accurate geometric models can be obtained by 3D scanners and material appearance can be defined by “Reflectance Distribution Function” such as “BSSRDF” and “BRDF”. And real world illumination conditions can be represented by HDRI (High Dynamic Range Image). But to exploit this information, it requires heavy computation time. So, in this paper we propose a method to estimate illumination conditions efficiently for realistic synthetic scene.
To analyze the distribution of real world illumination we exploit the segmented HDRIs which represent real world illumination. Rendering result with virtual objects using the proposed method is closely mimic the objects rendered using IBL method (Image Based Lighting) that is based on global illumination technique. Proposed method can be used some applications where computational time is vital factor, estimating few number of light sources. Also it is possible that use the lighting effect of each segmented image.
In the last few years, significant progress has been made in
the area of estimating the reflectance properties and the
illumination conditions of a scene based on images of the real
world scene. Initial research in this area started for shape from
shading application, which focused on recovering most
prominent single light source [
Figure1 shows the overall procedure for estimation of
lighting condition. For the first we need to segment input HDRI
as the intensity of radiation, it can be segmented many regions
but in our experiment we segment it into three regions; bright,
medium bright and dim region. Using these segmented images,
light sources are estimated by using median cut algorithm and
the result can be used rendering scene [
We use tone-mapped image which has LDR-level range of
the HDRI to segment the HDRI using the characteristics of
image histogram. We do this since it is hard to classify the
distribution of the pixel values due to the wide range of HDRI.
In addition, user intervention is required to determine the
characteristic value of the histogram. We use Fuzzy-C Means
clustering algorithm to automatically find a characteristic value
of the histogram without user intervention [
Figure 2 shows the procedure of FCM algorithm. The FCM algorithm is operated until the difference of the current and previous set is less than 0.0001. Then the light source information is computed from each segmented HDR image by using the median cut sampling algorithm. intensity radiation. Median Cut algorithm operates to partition an image into 2n regions of similar light energy.
We estimate the light sources based on the intensity of
radiation by using Median Cut algorithm [
Using the segmented HDR images, light sources can be estimated more efficiently such as more light sources are estimated in bright regions. Therefore it can generate realistic shadows using the small number of light sources. And it can be efficiently used in AR and interactive applications where realistic shadow is required and rendering time is a vital factor.
This research is funded by ETRI OCR and ICRC project at GIST.