An extensive empirical evaluation of focus measures for digital photography

Mir, Hashim; Xu, Peter; Beek, Peter van der

doi:10.1117/12.2042350

Cited by 33 publications

(27 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Connection with Shape-from-Focus and Autofocus: Several of the proposed loss functions have been proven successful in shape-from-focus (SFF) and autofocus (AF) with conventional cameras [43,44,51], showing that there is a strong connection between these topics and event-based motion estimation. The principle to solve the frame-based and event-based problems is the same: maximize a focus score of the considered image or histogram.…”

Section: Discussion Of the Focus Loss Functionsmentioning

confidence: 99%

Focus Is All You Need: Loss Functions for Event-Based Vision

Gallego

Gehrig

Scaramuzza

2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

113

116

View full text Add to dashboard Cite

θ I(x; θ) Parameters Warped Events Focus score Warping along point trajectories X [pix] Y [pix] time [s] X [pix] Y [pix] Measure event alignment Input Events θFigure 1: Motion Compensation Framework. Events in a space-time window are warped according to point trajectories described by motion parameters θ, resulting in an image of warped events (IWE) I(x; θ). Then, a focus loss function of I measures how well events are aligned along the point trajectories. This work proposes multiple focus loss functions for event alignment (last block in the figure and Table 1) for tasks such as rotational motion, depth and optical flow estimation. AbstractEvent cameras are novel vision sensors that output pixellevel brightness changes ("events") instead of traditional video frames. These asynchronous sensors offer several advantages over traditional cameras, such as, high temporal resolution, very high dynamic range, and no motion blur. To unlock the potential of such sensors, motion compensation methods have been recently proposed. We present a collection and taxonomy of twenty two objective functions to analyze event alignment in motion compensation approaches (Fig. 1). We call them focus loss functions since they have strong connections with functions used in traditional shapefrom-focus applications. The proposed loss functions allow bringing mature computer vision tools to the realm of event cameras. We compare the accuracy and runtime performance of all loss functions on a publicly available dataset, and conclude that the variance, the gradient and the Laplacian magnitudes are among the best loss functions. The applicability of the loss functions is shown on multiple tasks: rotational motion, depth and optical flow estimation. The proposed focus loss functions allow to unlock the outstanding properties of event cameras. †

show abstract

Section: Discussion Of the Focus Loss Functionsmentioning

confidence: 99%

Focus Is All You Need: Loss Functions for Event-Based Vision

Gallego

Gehrig

Scaramuzza

2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

113

116

View full text Add to dashboard Cite

show abstract

“…To solve this problem, we propose to measure image sharpness versus tomosynthesis plane position from a single non-contrast scan frame using a process that is similar to autofocus in digital photography. 12 It is expected that a curve describing tomosynthesis image sharpness vs. plane position (z) will exhibit maxima at z-planes corresponding to the edges of natural anatomical features, such as the vertebrae, ribs, diaphragm boundary, lung vasculature, and cardiomediastinal contour. Furthermore, the positions of these anatomic features are expected to be correlated to the position of the cardiac volume along the source-detector axis.…”

Section: Methodsmentioning

confidence: 99%

Localization of cardiac volume and patient features in inverse geometry x-ray fluoroscopy

Speidel

Slagowski

Dunkerley

et al. 2017

Medical Imaging 2017: Physics of Medical Imaging

View full text Add to dashboard Cite

The scanning-beam digital x-ray (SBDX) system is an inverse geometry x-ray fluoroscopy technology that performs real-time tomosynthesis at planes perpendicular to the source-detector axis. The live display is a composite image which portrays sharp features (e.g. coronary arteries) extracted from a 16 cm thick reconstruction volume. We present a method for automatically determining the position of the cardiac volume prior to acquisition of a coronary angiogram. In the algorithm, a single non-contrast frame is reconstructed over a 44 cm thickness using shift-and-add digital tomosynthesis. Gradient filtering is applied to each plane to emphasize features such as the cardiomediastinal contour, diaphragm, and lung texture, and then sharpness vs. plane position curves are generated. Three sharpness metrics were investigated: average gradient in the bright field, maximum gradient, and the number of normalized gradients exceeding 0.5. A model correlating the peak sharpness in a non-contrast frame and the midplane of the coronary arteries in a contrast-enhanced frame was established using 37 SBDX angiographic loops (64–136 kg human subjects, 0–30° cranial-caudal). The average gradient in the bright field (primarily lung) and the number of normalized gradients >0.5 each yielded peaks correlated to the coronary midplane. The rms deviation between the predicted and true midplane was 1.57 cm. For a 16 cm reconstruction volume and the 5.5–11.5 cm thick cardiac volumes in this study, midplane estimation errors of 2.25–5.25 cm were tolerable. Tomosynthesis-based localization of cardiac volume is feasible. This technique could be applied prior to coronary angiography, or to assist in isocentering the patient for rotational angiography.

show abstract

“…Many focus measures have been proposed and evaluated in the literature see, for example, [8,13]. In our work, we make use of two focus measures: (i) the squared gradient focus measure [17] and (ii) a focus measure based on the convolution of the image with the derivatives of the Gaussian [5,7].…”

Section: Focus Measuresmentioning

confidence: 99%

“…Both of these focus measures have been shown to be highly effective across many types of images [13] and the Gaussian focus measure has excellent noise reduction properties for low-light situations [5]. Following Kehtarnavaz and Oh [11], we assume that the region of interest (ROI) is the entire image.…”

Section: Focus Measuresmentioning

confidence: 99%