I. The Essence of Sharpness
Sharpness is a core metric for evaluating the detail rendition and edge clarity of an image. Its essence lies in the contrast gradient at the boundaries between regions of different tones or colors—the steeper the gradient, the crisper the edges and the stronger the detail separation. It not only determines the visual “perception of clarity” but also reflects the efficiency of image information transfer. In evaluation systems, it is typically quantified and analyzed using metrics such as Acutance and MTF (Modulation Transfer Function).

How to Distinguish “Sharpness”, “Resolution”, and “Clarity”?
Resolution is the hardware limit of “how much detail an imaging system can capture” (e.g., pixel density, line pairs per millimeter), serving as the foundation for detail rendition; Sharpness reflects “how crisp the detail edges are” and can be quantified by objective metrics like MTF50; Clarity is the subjective perception of the overall image sharpness by the human eye, resulting from the combined effects of resolution, sharpness, noise, contrast, and other factors.

II. Key Factors Affecting Sharpness
1. Hardware Level
Lens Performance: Design precision and manufacturing processes (such as lens coating quality and barrel light-shielding structures) directly impact edge contrast and stray light control; the aperture and focal length must be properly matched (large apertures tend to introduce aberrations like spherical aberration and coma, while medium/small apertures yield better sharpness; the telephoto end of a zoom lens typically performs better in sharpness than the wide-angle end); there are inherent differences across field positions, with the center region generally exhibiting higher sharpness than the peripheral regions (due to the distribution of lens aberrations).

Sensor Characteristics: The Anti-Aliasing Filter (AA filter, also known as a low-pass filter) slightly sacrifices objective sharpness to suppress aliasing artifacts such as moiré and false colors; sensors without an AA filter or with a weak AA filter are better at preserving sharpness but may introduce such aliasing artifacts.

Shooting Environment: Camera shake disrupts the spatial integrity of edges, leading to blurred details; focus accuracy is a prerequisite for sharpness (defocus directly causes overall or local edge blur); atmospheric turbulence (thermal convection, aerosol scattering) scatters light, reducing the original contrast of scene details and indirectly affecting image sharpness.

2. Signal Processing Level

Sharpening Algorithms: Enhances subjective sharpness by increasing the grayscale differences of edge pixels; however, over-sharpening can produce artifacts such as “halos” and “white fringes” at edges, degrading image quality; adaptive sharpening (e.g., bilateral filtering, guided filtering) can distinguish edges from noise based on pixel similarity, achieving a balance between sharpness and image quality by “enhancing edges while suppressing noise amplification”.

Noise Reduction Processing: High-frequency noise reduction algorithms (such as wavelet denoising) tend to mistakenly identify fine textures as noise and smooth them out when suppressing high-frequency noise, resulting in decreased sharpness; there is a trade-off between sharpening and noise reduction, requiring algorithmic optimizations such as regional processing and dynamic threshold adjustment to balance the effects of both.

III. Quantitative Metrics Related to Sharpness
1. Acutance
Definition: Acutance is an objective quantitative metric that closely aligns with the human eye's subjective perception of sharpness. Its results are strongly correlated with image viewing conditions: the same image may appear sharp on a small display but blurry when enlarged or viewed on a large display. This difference is determined by the contrast sensitivity characteristics of the human eye.

Principle: Following the method in ISO 12233:2023 Annex L, it uses the human Contrast Sensitivity Function (CSF) as a weighting basis to integrate objective Spatial Frequency Response (SFR) data, ultimately yielding a quantitative value (represented by the Q value) that correlates with subjective sharpness perception.
Its mathematical model is:
$$Q = \frac{\sum_{i=1}^{N} SFR_i \, CSF_i}{\sum_{i=1}^{N} CSF_i}, \quad i = 1, 2, \dots, N \tag{L.2}$$ where \(SFR_i\) is the Spatial Frequency Response value at the \(i\)-th spatial frequency; \(CSF_i\) is the human contrast sensitivity weight at the \(i\)-th spatial frequency; and \(N\) is the number of spatial frequency points involved in the calculation.

Contrast Sensitivity Function (CSF)
The core of Acutance is the Contrast Sensitivity Function (CSF), whose model originates from the S-CIELAB perceptual color space and has been adopted by the ISO 12233 standard for perceptual weighting of the luminance channel. The CSF curve (see figure below) describes the human eye's sensitivity to contrast at different spatial frequencies:
Horizontal Axis: Spatial frequency f, in units of cycles/degree, representing the fineness of image details.
Vertical Axis: Contrast sensitivity; a higher value indicates that the human eye can more easily discern contrast differences at that frequency.
The curve shows that the human eye is most sensitive to mid-to-low frequency details at around 4 cycles/degree, while its perceptual ability for both extremely high and extremely low frequency details drops significantly.
Its corresponding mathematical model is:
$$csf_{\text{lum}}(f) = \frac{(a \cdot f^c) e^{-bf}}{K} \tag{L.1}$$ where the parameter values are explicitly defined in ISO 12233:2023 Annex L: $a = 75$, the amplitude coefficient; $b = 0.2$, the high-frequency attenuation coefficient; $c = 0.8$, the frequency power coefficient; $K = 102.16$, the normalization constant (making the curve peak at 1.0); and $f$ is the spatial frequency in units of cycles/degree.

Viewing Conditions
Acutance results are strongly correlated with image viewing conditions: the same image may appear sharp on a small display but blurry when enlarged or viewed on a large display. To ensure that Acutance values calculated across different devices and laboratories are comparable, ISO 12233:2023 Annex L explicitly defines three standard viewing conditions (VC1/VC2/VC3) (see table below) to unify key parameters such as image display size, viewing distance, and pixel pitch. Before applying the CSF model to SFR data, the data units must first be converted from cycles per pixel to cycles per degree. According to the Nyquist sampling theorem, the maximum representable frequency $f_{cut} $ of an image is 0.5 cycles per pixel. To convert the frequency from “cycles per pixel” to “cycles per degree” for human contrast sensitivity calculations, the pixel size (i.e., pixel pitch p) and viewing distance D of the observer when viewing the final image must be known. This unit conversion is accomplished via Equation (L.3): $$f_{\text{cycles/degree}} = \frac{\pi D}{180 p} f_{\text{cycles/pixel}} = \frac{\pi D N_H}{180 H} f_{\text{cycles/pixel}} \tag{L.3}$$ where $N_H$ is the number of pixels in the vertical direction of the image; $H$ is the image height; $p$ is the pixel pitch of the display device; and $D$ is the viewing distance.

2. MTF/SFR
Definition: The Modulation Transfer Function (MTF) refers to the function of modulation varying with spatial frequency. It is essentially the same as SFR (Spatial Frequency Response), and the two terms can be used interchangeably.
Principle: By using slanted edges, bar patterns, or circular edges, it analyzes the attenuation of the amplitude (contrast) of a sinusoidal input as spatial frequency changes, quantifying the contrast transfer capability of the imaging system at different spatial frequencies.

Derived Metrics:
MTF50: The spatial frequency at which the MTF drops to 50%, which has the strongest correlation with visual sharpness and is the industry-standard metric.
MTF50P: The spatial frequency at which the MTF drops to 50% of its peak value; the peak may exceed 1 due to over-sharpening.
MTF10: The spatial frequency at which the MTF drops to 10%, close to the lower limit of human detail recognition, often used to evaluate “limiting resolution”.
Common Units: Cycles/Pixel, LW/PH (Line Widths per Picture Height)

See More
mtf_sfr_深度解析与测试实践