![](bg1.png)
Efficient Projective Transformation and Lanczos Interpolation on ARM
Platform using SIMD Instructions
Konstantinos Papadopoulos and Kyriakos Vlachos
Computer Engineering and Informatics Department, University of Patras, Patras, Greece
Keywords:
NEON, SIMD, Image Processing, Projective Transformation.
Abstract:
This paper proposes a novel way of exploiting NEON SIMD instructions for accelerating projective transfor-
mation in ARM platforms. Instead of applying data parallelism to linear algorithms, we study the effectiveness
of SIMD intrinsics on this non-linear algorithm. For image resampling, Lanczos interpolation is used since
it is adequately accurate, despite its rather large complexity. Multithreading is also employed for optimal use
of system resources. Moreover, qualitative and quantitative results of NEON’s performance are presented and
analyzed.
1 INTRODUCTION
Projective transformation is used in a wide range of
computer vision applications. It provides a linear
mapping between arbitrary quadrilaterals which is
very useful for deforming images controlled by mesh
partitioning. Some of the most well-known applicati-
ons are the removal of perspective distortion, image
stabilization, panoramic mosaic creation and object
tracking. Moreover, Lanczos resampling is one of
the most accurate algorithms for image upscaling, ac-
cording to (Burger and Burge, 2009). However, it is
computationally intensive, which can result in a poor
performance. The demand of faster multimedia appli-
cations is high, therefore improving projective trans-
formation’s processing time is crucial.
SIMD units’ contribution in multimedia applica-
tion development has been significant over the past
years. It allows parallel execution of both data type
operations (arithmetic, logical, etc.) and load/store
operations. Theoretically, this unit is able to accele-
rate operations up to 16 times, but this applies only
to certain data types. Optimal use of SIMD is possi-
ble at the low assembly level. However, developers
have the option to use SIMD intrinsics in high-level
programming (C/C++), taking advantage of interope-
rability and improved control over data.
Work presented in (Welch et al., 2012) regards
the implementation 2D bilinear interpolation algo-
rithm using NEON SIMD instructions. This algo-
rithm is exclusively used for image scaling. The
speedup achieved compared to the baseline algorithm
was 1.97-2.06 times. Moreover, in (Mitra et al.,
2013), authors proposed SIMD vector operations to
accelerate code performance on both low-powered
ARM and Intel platforms. They implemented Float
to Short data type conversion, binary image threshol-
ding, Gaussian Blur filter, Sobel filter and edge de-
tection algorithms in various ARM devices and ma-
naged to achieve speed gains from 1.05 to 13.88 com-
pared to compiler auto-vectorization. In addition,
(Mazza et al., 2014) achieved a speed gain of 3.76-
3.86 in bilinear interpolation using multithreading (2
Cortex-A9 cores) and SIMD instructions. Additional
work in SIMD multimedia processing field includes
linear image processing using OpenCL’s SIMD capa-
bilities in (Antao and Sousa, 2010) and acceleration of
alpha blending algorithm in a Flash application using
the Intel x86-64 platform’s SIMD (SSE) instructions
in (Perera et al., 2011).
This paper proposes a way of accelerating pro-
jective transformation using NEON SIMD instructi-
ons. The chosen resampling method is Lanczos in-
terpolation which is demanding and computationally
heavy, but produces notably results in terms of accu-
racy. Multithreading is utilized too, offering efficient
use of CPUs’ resources. Overall performance evalu-
ation of the proposed implementation is based on the
speed gains. Qualitative evaluation is also provided
for the output frames.
Papadopoulos, K. and Vlachos, K.
Efficient Projective Transformation and Lanczos Interpolation on ARM Platform using SIMD Instructions.
DOI: 10.5220/0006547000950100
In Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2018) - Volume 4: VISAPP, pages
95-100
ISBN: 978-989-758-290-5
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
95