Hardware Friendly Robust Synthetic Basis Feature Descriptor

Abstract

Finding corresponding image features between two images is often the first step for many computer vision algorithms. This paper introduces an improved synthetic basis feature descriptor algorithm that describes and compares image features in an efficient and discrete manner with rotation and scale invariance. It works by performing a number of similarity tests between the feature region surrounding the feature point and a predetermined number of synthetic basis images to generate a feature descriptor that uniquely describes the feature region. Features in two images are matched by comparing their descriptors. By only storing the similarity of the feature region to each synthetic basis image, the overall storage size is greatly reduced. In short, this new binary feature descriptor is designed to provide high feature matching accuracy with computational simplicity, relatively low resource usage, and a hardware friendly design for real-time vision applications. Experimental results show that our algorithm produces higher precision rates and larger number of correct matches than the original version and other mainstream algorithms and is a good alternative for common computer vision applications. Two applications that often have to cope with scaling and rotation variations are included in this work to demonstrate its performance. they perform well on intensity images and are robust against image variations, these algorithms require complex computations and large memory storage. These requirements are not desirable for resource-limited embedded applications and applications that require real-time performance.Several binary feature description algorithms have been developed to use simpler computation to provide descriptor with smaller size. Binary Robust Independent Elementary Features (BRIEF) [5] and the Binary Robust Invariant Scalable Key-points (BRISK) [6] are two representative and popular algorithms. A newer and improved version of BRIEF called ORB or rBRIEF has been developed [7] to cope with rotation invariance. These binary descriptors algorithms are sensitive to image variations and can have fewer accurate matches and a reduced match count compared to SIFT and SURF. A more detailed comparison of these algorithms is included in the next section.In our previous work, a new binary descriptor algorithm called SYnthetic BAsis (SYBA) descriptor was developed to obtain a higher percentage of correct matches [8]. SYBA performs binary comparisons between the binarized feature region and synthetic basis images to obtain a compact feature descriptor. SYBA's performance and size are comparable to other binary feature descriptors. However, like other binary feature descriptors, SYBA's feature point matching accuracy suffers under large amounts of image variations.In this paper, we introduce an improved version of SYBA called robust-SYBA (rSYBA). rSYBA maintains the same reduced descriptor size and computational complexity as the original SYBA while increasing the number of correct matches and matching precision, especiall...

Optimization and Implementation of Synthetic Basis Feature Descriptor on FPGA

“…SURF performed the best for the Graffiti sequence in which significant perspective changes cause sever rotation and scale variations. Previous work also includes several applications of SYBA based on a software implementation [13], including visual odometry drift reduction [21].…”

SR-SYBA: A Scale and Rotation Invariant Synthetic Basis Feature Descriptor with Low Memory Usage

“…While enjoying the computational simplicity, SYBA suffers from scaling and rotation variations. Effort has been made to address this challenge [18]. Similar to SIFT's approach, this new version of robust SYBA (rSYBA) resizes the FRI to different scales and uses the dominant gradient orientation to normalize the FRI orientation to achieve scale and rotation invariance.…”

“…We compared the proposed SR-SYBA feature descriptor with our previous work of rSYBA following the same experiment settings in [18]. Experiments were conducted on the BYU Scaling and Rotation Dataset [18], which contains an original picture and its eight scaled and rotated versions. We matched the scaled and rotated images to the original image.…”

Comprehensive Evaluation of Multispectral Image Registration Strategies in Heterogenous Agriculture Environment

“…The accuracy of SIFT has been established to be one of the highest among feature detector descriptor algorithms for scale and rotation variations. It also has greater accuracy for image rotation than other algorithms and has been determined to be the most precise algorithm across all geometric transformations [ 19 , 20 , 21 , 22 ]. This technique can be particularly useful for improving MS image registration and alignment while utilizing binary masks, which can be used to segment and extract specific regions of interest from MS images.…”