A Novel 2D To 3D Image Technique Based On Object-

Oriented Conversion

Introduction

•Proposed a novel 2D to 3D effect image conversion architecture to automatically converting a single 2D image into the left and right eye images.• After left and right eye images, we can see the left and right images on our left and right eyes individually on the 3D monitor.

Related Works• Depth perception We can use depth cues and some logic to estimate

the depth of objects. After we get the depth map, we can simulate the binocular parallax to let a single 2D image with depth information generate the left and right images. By using 3D monitor, observers see the different images in two eyes.

• Computed Image Depth algorithm1. Image depth computation process that computes

the image depth parameters with the contrast, the sharpness and the chrominance of the input images.

2. 3D image generation process that generates the 3D images according to the depth parameters.

Our proposed 2D To 3D Image Conversion Algorithm

Image Segmentation

• First part is preceding process: Due to the CCD camera, use SSR (Single Scale Retinex) to red

uce the light reflection• Second part is feature extraction and classified method: Choose (H,S,I) color space is more like human intuition and to

independently control intensity or chromatic component is easier.

Use a traditional clustering method ( the FCM algorithm). With FCM method, the pixels belonging to a valid class are clustered. A cluster is determined if the max value of the membership function is below the threshold T.

• Last part is region merging: The original image will be divided into many regions, each regio

n has the unique label number. But there will be an under-segmentation problem, we use connected component searching method.

Depth Extraction

Depth Extraction• Find zero-plane (balance surface): The image will employ mask operation method to calculate

variance each pixel. And the zero-plane which has the maximum variance in the picture.

• Give the depth value to each object: The depth value is calculated according to the distance between

the bottom of the object and the zero-plane. • Some rule: We estimate that it is a

horizontal or vertical object. Because we give the horizontal object a gradual depth and vertical object a fixed depth, the observer will feel more correctable on 3D display.

Shift Algorithm

(a) Camera shift capturing method, (b) Camera center capturing method.

Shift Algorithm

•Linear shift algorithm:On the right eye images, if the depth of pixel is higher than the zero-plane, the pixel will rightwards shift. On the contrast, if the depth of pixel is less than the zero-plane the pixel will leftwards shift. and the number of shift pixels is direct proportion to the distance between the depth of pixel and zero-plane

Shift Algorithm

•Binocular vision shift algorithm:The scene projects to the left and right eye images with an angle rotation. The diagram of simulating the right eye image is shown below. We assume that two eyes focus on the point whose vertical coordinate is half of the maximum depth of the scene and horizontal coordinate is half of the width. We can get the projection depth (D’) on the following formulas.

Shift Algorithm

Shift Algorithm

Shift Algorithm

Interpolation Holes Algorithm

1. First, we count the size of holes in horizontal direction.

2. If the size of hole is one pixel, we use the average of the left side and right side pixel to interpolate it.

3. If the size of hole is more than one pixel, we will mirror the holes with the side which is lower in depth map.

Results

(a) Original image, (b) after segmentation, (c) after size filter, (d) the depth map, (e) left eye image generating form linear shift algorithm, (f) right eye image generating from linear shift algorithm, (g) the final left eye image by using linear shift algorithm, (h) the final right eye image by using linear shift algorithm, (i) left eye image generating form binocular vision shift algorithm, (j) right eye image generating form binocular vision shift algorithm, (k) the final left eye image by using binocular vision shift algorithm, (l) the final right eye image by using binocular vision shift algorithm.

(a) Original image, (b) after segmentation, (c) after size filter, (d) the depth map, (e) left eye image generating form linear shift algorithm, (f) right eye image generating from linear shift algorithm, (g) the final left eye image by using linear shift algorithm, (h) the final right eye image by using linear shift algorithm, (i) left eye image generating form binocular vision shift algorithm, (j) right eye image generating form binocular vision shift algorithm, (k) the final left eye image by using binocular vision shift algorithm, (l) the final right eye image by using binocular vision shift algorithm.

Intoduction