Real-time High-Dynamic Range Image Design Based On Multi-lens

In the previous situation, the image’s quality will be lowered when the scene is shot under contrasting light intensity conditions, due to the limited dynamic range of the camera. If the shooting is under strong light conditions, the image will capture rich details in bright places and hides the shadow details; if the shooting is in low light environments, the image will highlight the dark details while hiding the light ones. In that case, image turns out to be either overexposed or underexposed.To solve this problem, camera with a high dynamic range(HDR) function is invented. To capture a good-quality image of a scene, numerous shootings are needed and synthesized, therefore, all parts of the image, light or dark, prove to be very apparent in strong contrast shooting conditions. However, there is still a weak point that it may take too long and the image cannot be captured in real time, since it is a necessity to shoot the same scene several times.The paper presents a new multi-lens imaging system, which can achieve a good HDR video play. Three identical lenses are applied to record the same scene: the left, the middle and the right. The mirrors of the three lenses are fixed in the same plane, the lenses parallel to each other and on the same line. The middle lens, with a weak exposure, observes the light details in the dark; the other two lens, with identical strong exposure, observe dark details in the light.First, take the image in the right lens as a reference one, make distortion correction and calibration of the three lens, and apply corner matching algorithm to find the matching corner of the left image and middle image with the reference.Second, with the matching corner and projection transformation algorithm, map the left and middle image to the reference image and will get a projection transformation image, which may lead to a fusion parallax while being fused with the reference. Work out the fusion parallax of the left image with the reference by the block matching algorithm, and the fusion parallax of the middle image with the reference according to the relationship between baseline distances.Finally, bilateral filter the Y components of the middle lens image and the reference image, combined with HDR algorithm, a video with HDR effect is obtained; as to highly-used modules with much calculation, optimize and accelerate with SSE, so to make HDR output video a smooth play.