Image Formation and Digital Representation

Introduction

Computer vision begins before an algorithm is applied. A scene must first be illuminated, projected through an optical system, recorded by a sensor, converted into numbers, and stored in a digital representation. The resulting image is therefore not the scene itself; it is a measurement shaped by geometry, optics, illumination, sensor characteristics, sampling, quantization, noise, and file encoding.

Understanding this imaging pipeline is essential for interpreting computer-vision results. A recognition model trained on sharp, well-exposed images may perform poorly when images are blurred, underexposed, compressed, or captured by a different camera. Many apparent algorithmic failures are actually failures of image formation or data representation.

This chapter develops the physical and mathematical foundations of image formation, beginning with the pinhole camera model and extending to lenses, illumination, sensing, digital sampling, color, noise, resolution, and image formats.

Learning Objectives

By the end of this chapter, you should be able to:

1. Explain how illumination, optics, and sensors produce an image.

2. Derive and apply the basic pinhole-camera projection equations.

3. Distinguish spatial sampling, quantization, resolution, and dynamic range.

4. Explain the origins and consequences of blur, distortion, and image noise.

5. Compare RGB, grayscale, HSV, and other common color representations.

6. Select suitable image file formats for computer-vision applications.

7. Analyze an imaging pipeline and identify likely sources of error.

1. The Imaging Pipeline

A simplified imaging pipeline is:

Scene geometry → illumination and reflectance → optics → sensor exposure → analog signal → sampling and quantization → color processing → file encoding.

A visible object reflects, transmits, or emits electromagnetic radiation. The camera collects a portion of that radiation. The lens controls how rays are mapped onto the sensor, while the sensor converts optical energy into an electrical signal. Electronic circuitry then samples the signal at discrete pixel locations and represents its amplitude with finite numerical precision.

At each stage, information may be transformed or lost. For example:

• insufficient illumination produces low signal-to-noise ratio;
• lens imperfections produce blur and geometric distortion;
• inadequate sampling produces aliasing;
• limited quantization produces banding;
• lossy compression may remove fine texture;
• incorrect color conversion may alter the meaning of pixel values.

A computer-vision system should therefore be designed with the imaging conditions in mind, rather than treating all images as interchangeable arrays.

2. Light, Illumination, and Reflectance

Visible light is electromagnetic radiation approximately spanning wavelengths from 400 to 700 nanometres, although the exact boundaries are not sharply defined. The spectral power distribution of a light source describes how much energy it emits at different wavelengths. Daylight, incandescent lamps, fluorescent lamps, and light-emitting diodes have different spectral distributions and therefore produce different colors in images.

The measured image intensity depends on both the illumination and the object. A simplified model for a matte surface is

\[

I(x,y) \approx L(x,y)R(x,y),

\]

where \(I\) is recorded intensity, \(L\) is incident illumination, and \(R\) is surface reflectance. This is an approximation: geometry, shadows, interreflections, sensor response, and optics also influence the measurement. The model is most useful for explaining why a dark image may result either from a dark object or from insufficient illumination.

Surface appearance depends on reflectance properties. Diffuse surfaces scatter light broadly, while specular surfaces produce view-dependent highlights. Transparent and translucent materials introduce transmission and subsurface scattering. These effects complicate object recognition because the same object can have different pixel values under different viewpoints and lighting conditions.

Exposure

Exposure is controlled primarily by aperture, shutter time, and sensor sensitivity. A larger aperture admits more light but generally reduces depth of field. A longer exposure gathers more photons but can produce motion blur. Increasing electronic gain or ISO makes the signal numerically larger but does not increase the number of photons collected; it often makes noise more visible.

For a moving camera or object, image displacement during exposure produces motion blur. Thus, exposure settings involve a trade-off among brightness, noise, depth of field, and temporal sharpness.

3. The Pinhole Camera Model

The pinhole model is the fundamental geometric model of perspective projection. It assumes a camera with a single optical centre and an infinitesimally small aperture. Although no practical camera is an ideal pinhole, the model accurately describes central projection and is the basis for camera calibration and three-dimensional vision.

Let the camera centre be the origin of a three-dimensional coordinate system. Let the image plane be at distance \(f\) from the camera centre. A 3D point \(P=(X,Y,Z)\), with \(Z>0\), projects to image coordinates \((x,y)\). Similar triangles give

\[

x=f\frac{X}{Z}, \qquad y=f\frac{Y}{Z}.

\]

Here, \(f\), \(X\), \(Y\), and \(Z\) must use consistent units. The coordinates \(x\) and \(y\) are physical distances on the image plane. The equations assume a camera aligned with the coordinate axes, no lens distortion, and a projection plane convention that produces positive image coordinates. A real camera may use a virtual image plane or a different sign convention; the geometric relationship remains the same in magnitude.

The ratios \(X/Z\) and \(Y/Z\) show perspective: objects farther from the camera appear smaller. The mapping is not linear in depth. A point twice as far away does not retain the same image size; its projected coordinates are approximately halved.

Homogeneous Coordinates and the Intrinsic Matrix

In computer vision, the pinhole projection is commonly written in homogeneous coordinates:

\[

\lambda\begin{bmatrix}u\\v\\1\end{bmatrix}

=K\begin{bmatrix}X\\Y\\Z\end{bmatrix},

\]

where \((u,v)\) are pixel coordinates, \(\lambda\) is a nonzero scale factor, and

\[

K=\begin{bmatrix}

f_x & s & c_x\\

0 & f_y & c_y\\

0 & 0 & 1

\end{bmatrix}.

\]

The intrinsic matrix contains camera parameters: \(f_x\) and \(f_y\) are focal lengths measured in pixels, \((c_x,c_y)\) is the principal point, and \(s\) is the skew parameter. In most modern cameras, pixel axes are nearly orthogonal and \(s\) is close to zero. Dividing the first two homogeneous coordinates by the third yields pixel coordinates.

If the world coordinate system differs from the camera coordinate system, an extrinsic transformation is required:

\[

\lambda\mathbf{p}=K[R\mid t]\mathbf{P}_w,

\]

where \(R\) is a 3-by-3 rotation matrix, \(t\) is a translation vector, \(\mathbf{P}_w\) is a world point in homogeneous form, and \(\mathbf{p}\) is its image point. This model assumes a rigid camera and a perspective camera with negligible distortion.

Worked Example: Perspective Projection

Suppose \(f=20\) mm and a point has camera coordinates \(X=30\) mm, \(Y=10\) mm, and \(Z=100\) mm. Then

\[

x=20(30/100)=6\text{ mm},\qquad y=20(10/100)=2\text{ mm}.

\]

The point therefore appears 6 mm to the right and 2 mm above the optical axis on the image plane. If the point moves to \(Z=200\) mm while \(X\) and \(Y\) remain unchanged, its coordinates become \(x=3\) mm and \(y=1\) mm. This illustrates perspective scaling.

4. Lenses and Practical Optics

A practical camera uses a lens rather than a pinhole. A lens gathers more light and permits shorter exposures, but introduces optical effects.

Focal Length and Field of View

Focal length is the distance, under the paraxial approximation, between the optical centre of a lens system and its image plane when focused at infinity. A longer focal length gives a narrower field of view; a shorter focal length gives a wider field of view.

For a sensor dimension \(d\) and focal length \(f\), the corresponding angular field of view is

\[

\theta=2\arctan\left(\frac{d}{2f}\right).

\]

This relation assumes a rectilinear lens and describes the field of view in one sensor dimension. Wide-angle and fisheye lenses may not obey the rectilinear model over their full field.

Aperture, Focus, and Depth of Field

The aperture controls the opening through which light enters. The f-number is defined as

\[

N=\frac{f}{D},

\]

where \(D\) is the effective aperture diameter. A smaller f-number represents a larger aperture. Aperture affects exposure, diffraction, and depth of field. Depth of field is the range of object distances that appear acceptably sharp, according to a chosen circle-of-confusion criterion. It is not an absolute property of a camera; it depends on sensor size, viewing conditions, magnification, focus distance, and the acceptable sharpness threshold.

Optical Aberrations and Distortion

Defocus blur occurs when rays from a point do not converge on the sensor plane. Spherical and chromatic aberration arise because real lenses do not focus all rays or wavelengths identically. Vignetting reduces brightness toward image corners.

Radial distortion changes the apparent geometry of straight lines. In a common camera model, normalized coordinates \((x,y)\) are transformed using \(r^2=x^2+y^2\):

\[

x_d=x(1+k_1r^2+k_2r^4+\cdots),

\]

\[

y_d=y(1+k_1r^2+k_2r^4+\cdots).

\]

Here \(k_1,k_2,\ldots\) are calibration coefficients. Positive or negative coefficients can produce barrel or pincushion distortion, depending on the convention and parameter values. Tangential distortion may result from lens and sensor misalignment. Calibration estimates these parameters so that images can be undistorted or geometric measurements can be corrected.

5. Image Sensors and Photometric Measurement

Digital cameras commonly use CCD or CMOS sensors. Both contain photosensitive elements that convert incident photons into charge. CMOS sensors now dominate many applications because they support fast readout and low power operation, although sensor architecture and performance vary widely.

A pixel measures light over a finite area and exposure interval. The sensor’s spectral sensitivity determines how different wavelengths contribute to its response. A monochrome sensor records one intensity measurement per pixel. A color camera usually uses a color filter array, such as a Bayer pattern, in which neighbouring sensor elements measure different color bands. Missing color components are estimated by demosaicing.

A simplified sensor measurement model is

\[

y=g\,q+n_r+n_s,

\]

where \(q\) is the number of photoelectrons generated by incoming light, \(g\) is electronic gain, \(n_r\) is read noise, and \(n_s\) represents other signal-dependent or systematic noise. The exact statistical distributions depend on the sensor and electronics.

Photon arrival is often modeled as Poisson-like, so shot-noise variance increases with the expected number of photons. Read noise is associated with electronic readout and may be approximately independent of signal. Dark current produces charge even without light and increases with temperature and exposure time. Fixed-pattern noise causes pixel-to-pixel response differences and can often be reduced by calibration.

Signal-to-Noise Ratio

Signal-to-noise ratio (SNR) compares useful signal magnitude with noise magnitude. Its precise definition depends on the measurement and noise model. For a photon-limited sensor with expected signal \(N\) photoelectrons and negligible other noise, the standard deviation of shot noise is approximately \(\sqrt{N}\), giving

\[

\mathrm{SNR}\approx\frac{N}{\sqrt{N}}=\sqrt{N}.

\]

This result applies to the idealized shot-noise-limited case. It shows why increasing illumination can improve image quality: the signal grows faster than its standard deviation. In a real camera, read noise, dark current, quantization, and saturation must also be considered.

6. Sampling and Spatial Resolution

A physical image varies continuously across space, but a digital image records samples at discrete locations. If the sampling interval is \(\Delta x\), the sampling frequency is \(f_s=1/\Delta x\) samples per unit distance. According to the Nyquist sampling principle, a band-limited signal must be sampled at more than twice its highest spatial frequency to avoid aliasing.

Aliasing occurs when fine scene patterns are represented as false lower-frequency patterns, such as jagged edges, moiré patterns, or changing wheel spokes in video. Real scenes are not perfectly band-limited, so cameras commonly use an optical low-pass filter or controlled defocus before sampling. Image resizing should also apply appropriate low-pass filtering before reducing resolution.

Spatial resolution is not identical to pixel count. It depends on pixel pitch, lens sharpness, focus, motion, sensor sampling, demosaicing, and image processing. A camera with more pixels does not necessarily resolve more detail if its lens or motion blur is limiting performance.

Temporal Sampling

Video also samples time. If motion changes too rapidly relative to the frame rate, temporal aliasing occurs. Shutter time and frame rate jointly determine motion representation. A high frame rate can reduce temporal aliasing, but only if sufficient light and suitable exposure are available.

7. Quantization and Dynamic Range

After sampling, each pixel value is quantized into one of a finite number of digital levels. An \(n\)-bit unsigned representation provides

\[

2^n

\]

possible code values. Thus, 8-bit data provide 256 levels, commonly numbered 0 through 255; 12-bit data provide 4096 levels.

Uniform quantization maps a continuous or high-precision signal to discrete intervals. Quantization error is the difference between the original value and its represented value. For an ideal uniform quantizer with step size \(\Delta\), and assuming error is uniformly distributed over \([-\Delta/2,\Delta/2]\), the quantization-noise variance is \(\Delta^2/12\). This approximation is most reasonable when the signal spans many quantization intervals and is not correlated with the quantizer.

Dynamic range is the ratio between the largest nonsaturating signal and the smallest useful signal. It is distinct from bit depth: increasing bit depth can represent levels more finely, but it does not automatically increase the sensor’s physical dynamic range. A sensor may clip bright regions at saturation while dark regions are buried in noise.

Linear sensor values are proportional, approximately, to measured light after suitable calibration. Display-oriented images often use a nonlinear transfer function such as sRGB gamma encoding. Confusing linear and nonlinear values can produce incorrect averaging, brightness calculations, and photometric analysis.

8. Grayscale and Color Representation

A grayscale image stores one intensity-like value per pixel. A color image stores multiple channels, but the meaning of each channel depends on the color space.

RGB

RGB represents color using red, green, and blue channels. It is convenient for displays and many image-processing libraries, but RGB values are device-dependent unless a color profile and a defined standard are supplied. Common standards include sRGB and wider-gamut spaces such as Adobe RGB or Display P3.

A grayscale conversion is often computed using a weighted sum rather than a simple average. For sRGB-like encoded values, a commonly used approximation is

\[

Y=0.299R+0.587G+0.114B.

\]

The coefficients reflect the different perceptual contributions of the channels. This formula is an engineering convention for a particular type of encoded RGB data, not a universal physical law. For accurate luminance calculations, RGB values should first be linearized and then transformed using the relevant colorimetry standard.

HSV and HSL

HSV and HSL separate a color-related angle or hue from saturation and a brightness/lightness-related component. They can be useful for segmentation when an object differs from its background in hue. However, hue becomes unstable when saturation is low, and these spaces are not perceptually uniform.

CIE and Perceptual Spaces

CIE XYZ provides a device-independent color representation based on standard observer functions. CIELAB was designed to provide a more perceptually organized space, although it is not perfectly uniform. Color spaces such as CIELAB and CIE LCh are useful for color differences, clustering, and analysis under controlled assumptions.

Chromaticity and YCbCr

Chromaticity representations separate color proportions from overall intensity. YCbCr separates a luma-like component from two chroma components and is widely used in video and JPEG-related workflows. Chroma subsampling, such as 4:2:0, reduces chroma spatial resolution because human vision is generally less sensitive to chroma detail than to brightness detail. This can be harmful for tasks requiring precise color boundaries.

9. Image Noise and Image Quality

Important image degradations include:

• Shot noise: variation caused by photon arrival statistics.
• Read noise: electronic noise introduced during sensor readout.
• Dark-current noise: thermally generated charge.
• Fixed-pattern noise: stable or slowly varying spatial response differences.
• Salt-and-pepper noise: isolated extreme-valued pixels, often associated with transmission or hardware faults.
• Quantization noise: error caused by finite digital precision.
• Motion blur: spatial integration of a moving scene during exposure.
• Defocus blur: spreading caused by incorrect focus.
• Compression artifacts: blocking, ringing, or loss of texture caused by lossy coding.

Noise reduction should respect the task. A smoothing filter may improve segmentation while destroying edges needed for recognition or measurement. Median filtering can suppress isolated impulse noise, whereas Gaussian filtering is often used for general smoothing. More advanced methods include non-local means, bilateral filtering, and learned denoisers. These methods may hallucinate or remove features, so evaluation should use task-relevant criteria rather than visual appearance alone.

10. Image File Formats and Data Management

An image format specifies how pixel data and metadata are stored. It may be uncompressed, losslessly compressed, or lossy.

• BMP: simple and often uncompressed; large files and limited metadata.
• PNG: lossless compression, transparency support, and suitable for diagrams, masks, and images requiring exact pixel preservation.
• JPEG: lossy compression based on transform coding; efficient for natural photographs but may introduce blocking, ringing, and color artifacts.
• TIFF: flexible container supporting high bit depths, metadata, and lossless or lossy compression; common in scientific, archival, and publishing workflows.
• GIF: limited color palette and lossless compression; suitable for simple graphics and animation, not high-quality continuous-tone imagery.
• RAW formats: preserve relatively direct sensor measurements, often with high bit depth and camera-specific metadata; require demosaicing and color processing.
• WebP and AVIF: modern formats offering efficient lossy and lossless compression, with support varying across software environments.

A file extension alone does not fully specify the data. Important properties include bit depth, channel order, color profile, alpha-channel meaning, chroma subsampling, compression settings, orientation metadata, and whether values are linear or gamma-encoded. For machine-learning datasets, images should be decoded consistently and metadata assumptions documented.

11. Applied Case: Vision Inspection on a Manufacturing Line

Background

Consider an illustrative manufacturing system that uses a camera to detect scratches on painted metal panels. The panels move beneath a fixed camera, and an automated system must identify defects without slowing production.

Problem

Initial tests show inconsistent results. Some scratches are detected, while others disappear when the production line operates at full speed. Dark panels are also more difficult to inspect than light panels.

Analysis

The first issue is motion blur. If the panel moves during a long exposure, a scratch is spread across several pixels and its contrast is reduced. The second issue is illumination: diffuse overhead lighting does not reveal shallow scratches reliably, while directional lighting can create useful highlights. The third issue is sensor noise in dark regions. Finally, JPEG compression is found to introduce artifacts around thin high-contrast defects.

Approach and Solution

The team uses short exposure times and synchronized strobed illumination to freeze motion. Lighting is placed at a shallow angle to increase sensitivity to surface irregularities, while additional diffuse lighting reduces excessive specular variation. The camera is calibrated to correct lens distortion and is mounted rigidly to prevent changes in viewpoint. Images are captured in a lossless format during development, with sufficient bit depth to preserve weak intensity differences. A preprocessing stage performs flat-field correction to compensate for vignetting and fixed-pattern response. The inspection model is trained using images collected under the intended operating conditions, including different panel colors and acceptable manufacturing variation.

Outcome

Defect detection becomes more stable because the image formation process is controlled before algorithmic classification. The system also records exposure, illumination, camera temperature, and calibration version, making failures easier to diagnose.

Learning Points

1. Better lighting can be more valuable than a more complex neural network.

2. Motion blur and compression can remove information permanently.

3. Calibration and metadata are part of a reliable vision system.

4. Training data must represent the physical conditions under which deployment occurs.

5. Image quality should be measured with respect to the inspection task, not only by visual appeal.

12. Professional Practice, Ethics, and Contemporary Developments

Camera selection should consider field of view, working distance, sensor size, spectral sensitivity, frame rate, exposure control, dynamic range, environmental protection, and synchronization. In safety-critical applications, the full measurement chain should be validated, not merely the classifier.

Privacy is a major governance concern. Cameras may collect faces, license plates, biometric information, or activities unrelated to the stated purpose. Responsible systems should apply data minimization, access control, retention limits, appropriate consent or legal basis, and documented impact assessments. Image enhancement must not be presented as recovery of information that was never captured. Generative restoration may create plausible but false details, which is especially dangerous in forensic, medical, and legal contexts.

Contemporary developments include event-based sensors, high-dynamic-range imaging, multispectral and hyperspectral cameras, computational photography, global-shutter CMOS sensors, and learned image formation pipelines. Event cameras record asynchronous brightness changes rather than conventional intensity frames, offering high temporal resolution but requiring different algorithms. Multispectral systems extend sensing beyond ordinary RGB, but their calibration and data-management requirements are more demanding.

Best practice is to document the camera model, lens, calibration, lighting, exposure, color space, bit depth, preprocessing, compression, and train-test acquisition conditions. Reproducibility depends on these details.

Chapter Summary

Image formation is a physical and computational process linking a three-dimensional scene to a digital array. Illumination and reflectance determine available visual information; lenses project and modify it; sensors convert photons into electrical measurements; sampling discretizes space and time; quantization limits numerical precision; and file formats encode the result.

The pinhole model explains perspective projection through \(x=fX/Z\) and \(y=fY/Z\), while the intrinsic matrix expresses focal lengths, principal point, and skew in pixel coordinates. Real cameras introduce blur, distortion, vignetting, noise, and color-processing effects. Adequate sampling reduces aliasing, while suitable bit depth and dynamic range preserve tonal information. RGB, HSV, YCbCr, and perceptual spaces serve different purposes, and file-format choices affect both fidelity and storage.

For computer-vision practice, image quality is not independent of the task. Controlled illumination, appropriate exposure, calibration, loss-aware storage, and documented preprocessing often determine whether an AI system is reliable. Ethical practice additionally requires careful handling of personal imagery and honest communication about what image processing can and cannot recover.

Key Terms

Pinhole camera: An ideal central-projection camera with an infinitesimally small aperture.

Perspective projection: Mapping in which image size varies inversely with depth.

Focal length: Optical distance associated with focusing parallel rays at the image plane.

Principal point: Image location corresponding approximately to the optical axis.

Intrinsic parameters: Camera parameters describing internal projection geometry.

Extrinsic parameters: Camera pose parameters describing rotation and translation relative to a world frame.

Aperture: Opening that controls the amount of light entering a lens.

Depth of field: Range of object distances appearing acceptably sharp.

Distortion: Geometric deviation from the ideal projection model.

Vignetting: Reduction in brightness toward the image periphery.

Sampling: Measurement of a continuous signal at discrete spatial or temporal locations.

Aliasing: False structure caused by insufficient sampling.

Quantization: Mapping continuous or high-precision values to discrete numerical levels.

Dynamic range: Ratio between the largest useful signal and the smallest useful signal.

Shot noise: Photon-counting variation associated with light arrival statistics.

SNR: Ratio comparing signal magnitude with noise magnitude.

Color space: Defined coordinate system for representing color information.

Demosaicing: Estimation of missing color components in a color-filter-array image.

Chroma subsampling: Reduction of chroma spatial resolution to save storage or bandwidth.

RAW image: Sensor-oriented data retaining substantial unprocessed measurement information.

Review Questions

Short Answer Questions

1. What physical factors determine the intensity recorded from a surface?

2. State the two basic perspective-projection equations in the pinhole camera model.

3. What is aliasing, and why can an optical low-pass filter help prevent it?

4. Distinguish bit depth from dynamic range.

5. Why can JPEG compression be unsuitable for pixel-accurate defect detection?

Long Answer Questions

1. Explain the complete imaging pipeline from illumination to image-file storage, identifying possible information losses at each stage.

2. Derive the pinhole projection equations using similar triangles and discuss the assumptions of the model.

3. Compare RGB, HSV, YCbCr, and CIELAB in terms of representation and applications.

4. Explain the main sources of sensor noise and discuss how exposure affects signal-to-noise ratio.

5. Discuss how lens properties, focus, aperture, and motion influence image quality in a computer-vision system.

Analytical / Application Questions

1. A camera has focal length 12 mm and a sensor width of 24 mm. Calculate its horizontal field of view using \(2\arctan(d/(2f))\), and explain what happens to the field of view if focal length is increased to 24 mm.

2. Design an imaging pipeline for inspecting rapidly moving objects. Justify your choices of lighting, exposure, sensor, file format, calibration, and preprocessing.

3. An object-recognition system performs well in the laboratory but poorly outdoors. Analyze at least four possible image-formation causes and propose controlled experiments to distinguish them.

Multiple Choice Questions with Answers

Question 1

Which equation represents horizontal perspective projection in the ideal pinhole camera model?

A. x=fX/Z

B. x=fZ/X

C. x=X+fZ

D. x=Z/(fX)

Correct Answer: A. x=fX/Z

Question 2

In the camera intrinsic matrix, which parameters locate the principal point?

A. fx and fy

B. cx and cy

C. R and t

D. k1 and k2

Correct Answer: B. cx and cy

Question 3

What is the main consequence of insufficient spatial sampling?

A. Vignetting

B. Chromatic aberration

C. Aliasing

D. Increased depth of field

Correct Answer: C. Aliasing

Question 4

How many code values are available in an ideal 10-bit unsigned image channel?

A. 10

B. 100

C. 512

D. 1024

Correct Answer: D. 1024

Question 5

Under an ideal shot-noise-limited model with N detected photoelectrons, the approximate SNR is:

A. N

B. 1/N

C. sqrt(N)

D. N squared

Correct Answer: C. sqrt(N)

Question 6

Which format is generally most appropriate when exact pixel preservation and lossless compression are required?

A. JPEG

B. PNG

C. GIF with a reduced palette

D. Low-quality WebP

Correct Answer: B. PNG

Question 7

What does a longer focal length generally do, for a fixed sensor size?

A. Widens the field of view

B. Narrows the field of view

C. Removes all lens distortion

D. Eliminates motion blur

Correct Answer: B. Narrows the field of view

Question 8

Which effect is most directly caused by object motion during exposure?

A. Motion blur

B. Quantization

C. Demosaicing

D. Chroma subsampling

Correct Answer: A. Motion blur

Question 9

Why is hue often unreliable for a nearly gray pixel in HSV?

A. The pixel has excessive dynamic range

B. Hue becomes unstable when saturation is low

C. HSV contains no brightness component

D. Gray pixels cannot be sampled

Correct Answer: B. Hue becomes unstable when saturation is low

Question 10

A 20 mm lens is used with a 20 mm-wide sensor. What is the approximate horizontal field of view?

A. 26.6 degrees

B. 45 degrees

C. 53.1 degrees

D. 90 degrees

Correct Answer: C. 53.1 degrees