Ramazan Duvar

Fuzzy fusion based high dynamic range imaging using adaptive histogram separation

In this work, a high dynamic range (HDR) image generation method using a single input image is presented. The proposed approach generates over- and under-exposed images by making use of a novel adaptive histogram separation scheme. Thus, it becomes possible to eliminate ghosting effects which generally occur when several input image containing camera/object motion are utilized in HDR imaging. Additionally, it is proposed to utilize a fuzzy logic based approach at the fusion stage which takes visibility of the inputs pixels into account. Since the proposed approach is computationally light-weight, it is possible to implement it on mobile devices such as smart phones and compact cameras. Experimental results show that the proposed approach is able to provide ghost-free and improved HDR performance compared to the existing methods.

Real-time implementation of 1-dimensional projection based video stabilization on a mobile device

In this work, a low complexity image stabilization method is implemented in real-time on a smart phone that has an ARM Cortex-A8 processor. In motion estimation process, initially projection (sum) of image frames along the rows and columns is computed in order to eliminate computational complexity of 2-D full search based approaches. Next, the searching process is performed by making use of the projection matrix of rows and columns. Finally, Kalman filter is used for motion correction which is commonly used in literature. The proposed method is executed at 15fps on Samsung Galaxy S smart phone which have 1.0GHz processor and Android 2.2 operating system, using Java Native Interface (JNI).

Multi band pass filter based one-bit transforms in motion estimation

One-bit transform method can be used for low computational complexity block based motion estimation with respect to hardware implementation in video coding systems. The principal of the one-bit transform method is based on kernel which has multi band pass filtering. In this work, the performance of the one-bit transform methods obtained with band pass filtering on the state of the art video coding standard has been evaluated. Also new kernels, which have different multi band pass filtering, have been proposed to compare the performance of the low bit representation based motion estimation in low and high resolution videos.

Implementation of 1 -bit transform based motion estimation to HEVC encoder

In this work, one-bit transform (1BT) based motion estimation method is applied to HEVC encoder which is the most recent video coding standard. The most significant portion of the computational load of the HEVC originates from the motion estimation. Thus, many approaches are presented in literature in order to reduce this computational load. 1BT based ME approach denotes input image frames in low-bit depth form so that it is not only suitable for hardware implementation but also reduces computational load of whole process since it uses EX-OR based matching criterion. In this work, 1BT based ME is integrated into HEVC encoder and its performance is assessed. Experimental results show that the 1BT based ME approach has a neglectable performance loss compared conventional HEVC motion estimation approach.

Real-time lane marking detection using modified 1-bit transform based pre-processing

In this work, an image processing based lane-detection approach is proposed. In the proposed approach, candidate pixels that can be used for lane markings are detected by making use of 1-bit transform as a pre-processing step. Next, feature points are extracted via Sobel filter and candidate lane markings are decided employing a correlation and Hough transform based approach. Finally, Kalman filter is utilized to exploit temporal correlation of lane markings in successive frames. Experimental results show that the proposed approach provides similar or better lane detection performance compared to some recently presented methods while having real-time processing capability.

Performance evaluation of low bit-dept based motion estimation approaches on screen content video

Screen contents have various characteristics features different from the natural images captured by cameras. These characteristic features have important effects on the video coding performance of motion estimation approaches. In this work, we carry out performance comparison of some low bit-depth based motion estimation approaches and evaluate performance of methods based on the characteristic features of screen content.

High dynamic range imaging on a mobile device using single input image

In this work, we present a high dynamic range (HDR) image generation method using a single input image that can be executed on a mobile device. The utilized method eliminates ghosting effect inherently occurring in HDR images when several input image containing camera/object motion are employed. The employed adaptive histogram separation scheme enables efficient low dynamic range image generation from a single input image with different characteristic and hence provides improved HDR performance. Since the utilized approach is computationally lightweight, it is possible to implement it on mobile devices such as smart phones and compact cameras. It is shown that the empolyed approach provides ghost-free and improved HDR performance compared to existing methods at experimental results. The HDR method is implemented on Samsung Galaxy Note 3 smart phone using Java Native Interface (JNI) efficiently.

Real time control of tyre profiles using computer vision approach

In this work, an online three-dimensional profile measurement systems in order to determine reasons, internal tensions etc., which affect tyre uniformity, balance and performance has been developed. Measurement system that was developed consists of two line laser and two cameras. Three dimensional model of the tyre profile has been reconstructed by projection of the line laser onto the tyre surface. The noises on the captured images has been removed by applying hourglass filter. After camera calibration has been done using calibration object, the three dimensional model has been reconstructed. The measurement system developed, is able to obtain 3-dimensional model of the tyre profile with 0.3 mm accuracy. After 3-dimensional model is obtained, a graphical user interface is developed in order to monitoring three dimensional mesh and measurements taken on the tyre.