“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

From the Publisher: 
The accessible presentation of this book gives both a general view of the entire computer vision enterprise and also offers sufficient detail to be able to build useful applications. Users learn techniques that have proven to be useful by first-hand experience and a wide range of mathematical methods. A CD-ROM with every copy of the text contains source code for programming practice, color images, and illustrative movies. Comprehensive and up-to-date, this book includes essential topics that either reflect practical significance or are of theoretical importance. Topics are discussed in substantial and increasing depth. Application surveys describe numerous important application areas such as image based rendering and digital libraries. Many important algorithms broken down and illustrated in pseudo code. Appropriate for use by engineers as a comprehensive reference to the computer vision enterprise.

Computer vision : a modern approach = 计算机视觉 : 一种现代的方法

MUCKE aims to mine a large volume of images, to structure them conceptually and to use this conceptual structuring in order to improve large-scale image retrieval. The last decade witnessed important progress concerning low-level image representations. However, there are a number problems which need to be solved in order to unleash the full potential of image mining in applications. The central problem with low-level representations is the mismatch between them and the human interpretation of image content. This problem can be instantiated, for instance, by the incapability of existing descriptors to capture spatial relationships between the concepts represented or by their incapability to convey an explanation of why two images are similar in a content-based image retrieval framework. We start by assessing existing local descriptors for image classification and by proposing to use co-occurrence matrices to better capture spatial relationships in images. The main focus in MUCKE is on cleaning large scale Web image corpora and on proposing image representations which are closer to the human interpretation of images. Consequently, we introduce methods which tackle these two problems and compare results to state of the art methods. Note: some aspects of this deliverable are withheld at this time as they are pending review. Please contact the authors for a preview.

http://ieeexplore.ieee.org/iel2/4524/12820/00592460.pdf

Image Processing

Review of shape representation and description techniques

50 Years of Image Analysis

In an autonomous driving assistance system (ADAS), top views effectively represent objects around the vehicle on a 2D plane. Top-view images are therefore widely used to detect lines in ADAS applications such as lane-keeping assistance and parking assistance. Because line detection is a crucial step for these applications, the false positive detection of lines can lead to failure of the system. Specular reflections from a glossy surface are often the cause of false positives, and since certain specular patterns resemble actual lines in the top-view image, their presence induces false positive lines. Incorrect positions of the lines or parking stalls can thus be obtained. To alleviate this problem, we propose two methods to estimate specular pixels in the top-view image. The methods use a geometric property of the specular region: the shape of the specular region is stretched long in the direction of the camera as the distance between the camera and the light source becomes distant, resulting in a straight line. This property can be used to distinguish the specular region in images. One estimates the pixel-wise probability of the specularity using gradient vectors obtained from an edge detector and the other estimates specularity using the line equation of each line segment obtained by line detection. To evaluate the performance of the proposed method, we added our methods as a pre-processing step to existing parking stall detection methods and investigated changes in their performance. The proposed methods improved line detection performance by accurately estimating specular components in the top-view images.

Specular Detection on Glossy Surface Using Geometric Characteristics of Specularity in Top-View Images.

Background subtraction, which separates moving objects in video sequences, is an essential technology for object recognition and tracking. However, background subtraction methods are often confused by shadow regions and this misclassification of shadow regions disturbs further processes to perceive the shapes or exact positions of moving objects. This paper proposes a method for shadow elimination which is based on shadow modeling by color information and Bayesian classification framework. Also, because of dynamic update of modeling parametres, the proposed method is able to correspond adaptively to illumination changes. Experimental results proved that the proposed method can eliminate shadow regions effectively even for circumstances with varying lighting condition.

An Effective Shadow Elimination Method Using Adaptive Parameters Update

Motion estimation has been a popular choice in video compression standards such as MPEG-1, MPEG-2, and MPEG-4 to eliminate temporal redundancy of video sequences. The motion estimation and motion compensation (MEMC) part in video encoding represents around 60–80 percent of the total computation, where MEMC is a core part that cannot be avoided for preserving substantial compression efficiency. In this paper, we have presented a method to minimize the use of MEMC, which will further reduce the computational complexity with the reasonable visual quality. The main idea of our proposal is to unselect ME on some frames with very low motion. Experimental results indicated that the proposed coding scheme can reduce about 55–75 percent in computational complexity over the MPEG-4 video reference software.

Selective motion estimation for fast video encoding

We introduce a method for the efficient calibration of around-view-monitoring (AVM) cameras. Particularly, we introduce two situations that require calibration because of the characteristics of AVM cameras: a situation wherein cameras are shipped from the manufacturing line and another situation wherein some cameras are distorted during operation and need recalibration. In this study, the calibration method for shipped cameras is defined as the factory mode and that for recalibration is defined as the service mode. In the factory mode, two circular patterns placed at a regular distance are used to ensure the maximum accuracy while requiring minimum calibration. In the service mode, as a recalibration method, we developed a robust method that considers various environments using parallel parking lines that can be easily installed in general service centers. In the factory mode, we confirmed that the AVM-camera-calibration error was within 5.66 cm when the two circular patterns were located at a certain distance within a certain range. However, in the service mode, we achieved camera movement angle error equal to or less than 0.1° using the parking-line-detection result.

/pdf/an-automatic-calibration-method-for-avm-cameras-24srw5h7ai.pdf

An Automatic Calibration Method for AVM Cameras

This paper proposes an algorithm for parking assist system. It detects a parking slot and estimates the pose of the parking slot. User touches the parking slot that wants to park at the display showing rear view. Then, an ROI is set based on the touch point. Parking slot is detected in the ROI using Hough transform. NCC (Normalized Cross Correlation) is used to get the pose of parking slot. It successfully detects parking slot and estimates the pose. But, there is a problem due to the scene change causing variation of parking slot at rear view when vehicle enters the parking slot. To solve this problem, the process of pose estimation is separated into phase 1 and phase 2 according to applied template shape.

Whoi-Yul Kim

Papers

Specular Detection on Glossy Surface Using Geometric Characteristics of Specularity in Top-View Images.

An Effective Shadow Elimination Method Using Adaptive Parameters Update

Selective motion estimation for fast video encoding

An Automatic Calibration Method for AVM Cameras

A method of detecting parking slot in hough space and pose estimation using rear view image for autonomous parking system