Understanding Digital Camera Resolution
By: JR GEOFFRION
Like the digital audio revolution of the mid-80s, photography and the darkroom are going digital.
Resolution is the most talked about digital camera characteristic and is often used to describe image quality. Yet, the fundamental lack of accurate information surrounding resolution makes it one of the most misunderstood discussion topic.
The issues of digital camera resolution are detailed throughout this article using laymen’s terms that can be understood by all. First, terminology and definitions are clarified. Second, concepts and design considerations such as pixel shape, size, and configuration are explained and introduced gradually.
In order not to overwhelm the reader, widely spread misconceptions are introduced and addressed in a progressive manner that build on each other. These misconceptions include:
· All [insert number] megapixel cameras have the same number of pixels
· Two cameras of equivalent megapixels can store the same number of pictures per megabyte
· Resolution is expressed in megapixels
· The resolving power of a 100 pixels per millimeter sensor is 50 line pairs per millimeter
The article concludes by making specific recommendations on how to assess and compare digital camera resolution.
Misconception #1: All [insert number] megapixel cameras have the same number of pixels
Lack of standard nomenclature is the leading contributor to confusion in this area. When a particular digital camera is referred to as having 3 megapixels, it may actually use a lot less than 3 megapixels to capture the picture.
How can this be? As described below, there are many definitions of pixel count.
As its name indicates, the total pixels number refers to the overall number of pixels on the sensor. However, this number is, more often than not, irrelevant and misleading due to the fact that not all pixels are illuminated by the image circle produced by the lens and used to capture the picture.
Some of the pixels on the sensor are simply not used at all while others are used for picture processing, rather than picture capture. Figure 1 depicts the picture area of the sensor that is illuminated by the image circle produced by the lens and used to capture the picture. The pixel count in this picture area is called the picture pixels and is sometimes also referred to as image pixels. Since picture pixels are the only pixels used to capture the picture, the picture pixels number is a more accurate indicator of a digital camera’s native resolution than is the number of total pixels – but not an accurate indicator for reasons that are later described.
Often, overall sensor dimensions (hence total pixels) are compared between various camera models. However, it is not the overall sensor dimensions that should be compared but rather the picture areas. The picture area size is derived by multiplying the individual pixel size (total pixels on the axis ÷ dimension of that axis) by the number of picture pixels on that axis.
As illustrated in figure 2, in some circumstances the image circle may be smaller than that of the picture area of the sensor. In this scenario, the number of picture pixels is much smaller than the total pixels number due to the un-maximized use of the picture area. This is one of the reasons why total pixels number can be misleading.
The reverse, where the image circle is much larger than the picture area of the sensor, is also possible. Common examples of this include professional digital cameras such as Canon’s D30 and EOS-1D as well as Nikon’s D1 series cameras that use lenses designed for 35mm film photography. In these cameras, the picture area does not cover the full 35mm frame of 24 by 36 millimeters. The ratio of picture pixel area diagonal to the diagonal of 35mm film gives rise to the so called “focal length multiplier” illustrated in Figure 3.
The number of effective pixels differs from picture pixels in that it also includes the pixels used to calibrate the black. This is accomplished by getting a ground or zero reading from masked pixels not exposed to light.
To ensure broad and general confusion, the picture pixels definition proposed by the Photographic and Imaging Manufacturers Association (PIMA) (now the International Imaging Industry Association) in their working draft of “PIMA 7000 – Photography – Digital Still Cameras – Guidelines for Reporting Pixel-Related Specifications” (and used in this article to alleviate any potential terminology confusion) is the same as the Japan Camera Industry Association’s (JCIA) definition of effective pixels.
Up until now, the three definitions above refer to pixels physically located on the sensor. However, for several reasons later explained, each picture pixel information must be interpolated to create a viewable image. The number of pixels in this stored viewable image is the recorded pixel number. Most cameras have settings allowing users to store images of different sizes, essentially changing the number of recorded pixels.
In certain instances, the number of pixels in the output image differs from that of the recorded pixels. Typically, this is due to cropping of the recorded image to fit standard dimensions such as VGA, SVGA, XGA, etc. or to fit aspect ratios such as 3:2, 4:3, and 16:9. Output pixels are sometimes also referred to as image pixels.
Misconception #1 Recommendation
Don’t be fooled by the total pixel
number. Rather, look at picture pixels (also referred to as effective pixels)
and output pixels to get a general idea of capabilities and limitations. However,
take these numbers with a grain of salt as will be explained later.
Misconception #2: Two cameras of equivalent megapixels can store the same number of pictures per megabyte
The restrictive factor for storage capacity is the recorded image file size, not dimensions. Unfortunately, there is no one-to-one relationship between image dimensions and file size. In addition to image dimensions, several other factors affect file size, most importantly: the picture itself, bit-depth, file formats, compression algorithms and settings, and additional metadata recorded (such as camera and user information, exposure mode, metering mode, date, time, file format, compression, image size, lens, focal length, focus distance, aperture, speed, white balance, flash output, ISO, etc.). Of these factors, compression algorithms and settings will affect resolution drastically. Why?
Not all compression algorithms are lossless. This means that information is lost during image compression and cannot be reconstructed during decompression. For example, the popular JPEG format does not use a lossless compression algorithm and loses image information when images are compressed. On the other hand, formats like the Tagged Image File Format (tiff) can be compressed without any image degradation. Raw format can sometimes be preferable, as the camera processor does not correct and modify the image. The downside of such a format is that it typically requires proprietary software to extract the image.
Misconception #2 Recommendation
Get ample storage space to capture
images in a lossless format. Increasing compression using a lossy algorithm
will degrade the image.
Misconception #3: Resolution is expressed in megapixels
For the purpose of this article, resolution is defined as the ability to distinguish (resolve) separate visual information such as details and fine patterns. Traditionally, the measure of resolution is described by the resolving power and is expressed in line pairs per millimeter (lp/mm). It is evaluated from the ability to distinguish distinct line patterns from standard test charts. However, resolving power numbers (lp/mm) are a somewhat subjective measure because each test administrator applies his or her own judgment in identifying at which point a pattern can no longer be distinguished. For this reason, no conclusions can be drawn from resolution figures comparison from various sources or derived from different methods but comparisons made under the same conditions and evaluated by the same person do offer a comparison reference. It should be noted however, that computerized methods to evaluate the cut-off point of a distinguishable pattern are currently being tested and proposed.
Images are two-dimensional objects. Therefore, doubling the resolving power necessitates a fourfold increase in pixel (of equal quality) – not a twofold increase as is commonly believed. All else being equal, a 6 megapixel camera does not have twice the resolving power of a 3 megapixel camera, but twice that of a 1.5 megapixel camera.
To complicate matters further, several other factors affect resolution. Some of the most important factors include sensor design, interpolation algorithm, the lens itself, focal length (in the case of zoom lenses), focus distance, aperture, position in the image field, orientation (horizontal, vertical, and diagonal), scene contrast, and vibrations. Due to the number of degrees of freedom, comparing two camera’s megapixels rating would be absolutely pointless as described later.
Misconception #3 Recommendation
Look way beyond the megapixel numbers.
Megapixel numbers are only an order of magnitude indicator of the camera’s resolution.
As it will be described later, performing standard tests is the only way of
assessing the camera’s true resolution capabilities.
Misconception #4: The resolving power of a 100 pixels per millimeter sensor is 50 line pairs per millimeter
There are a few main reasons why the above statement is not exact, namely:
· Not all sensor pixels are square
· Not all sensor pixel arrays are configured in a grid
· Aliasing may occur
· Contrast affects resolution
· Photodiode have a limited dynamic range
· Photodiodes are not sensitive to colors
As it will be seen, these reasons all stem from the fact that there is not a one-to-one relationship between sensor pixels and image pixels. If there was a one-to-one relationship, the horizontal resolving power of the simplified bar pattern in Figure 4 would be 1/p where p is twice the pixel-to-pixel distance in millimeter (since patterns are created of one black and one white line). This frequency is called the Nyquist frequency.
An important fact is often neglected when analyzing resolution. In a grid array, the diagonal pixel-to-pixel distance is greater than the horizontal or vertical distance by a factor of 1.41 (from Pythagoras’ theorem, the diagonal distance is [12+12]0.5 which is the square root of two). This reduces the resolving power in the diagonal direction.
Moreover, not all pixels are square. For example, Nikon uses rectangular pixels in its D1x with twice the horizontal pixel count than in their D1h. Hence, as illustrated in Figure 5, the horizontal pixel-to-pixel distance (ph) is half that of the vertical pixel-to-pixel distance (pv). In this case, the horizontal resolving power of the sensor is double that of its vertical resolving power. The D1x on-board processor redistributes this increased horizontal resolution, both horizontally and vertically, using proprietary algorithms.
Pixels are not necessarily aligned in a grid. For example, Fujifilm has developed a sensor using octagonal pixels configured in an interleaved array. This shape and configuration has allowed for a drastically shorter pixel-to-pixel distance in the diagonal orientation while increasing the photodiode size.
Primarily for these reasons and the fact that certain compression algorithms compress horizontal and vertical patterns differently, resolving power should be evaluated horizontally, vertically, and diagonally.
Human Visual System
It is interesting to note that, as demonstrated by Watanabe et al in 1968, our visual system is more sensitive to horizontal and vertical patterns than diagonal ones. This is fortunate as average scenes contain more horizontal and vertical patterns than diagonal ones.
Lenses do not transmit light with one hundred percent efficiency. As light enters the lens, a degradation of contrast occurs. This degradation is typically more visible at a lens’ full aperture, with wide angles, or in image corners. In some situations, degradation is such that vignetting occurs. However, no lens is immune to this phenomenon.
Hence, a low contrast pattern may not be transmitted through the lens and the image projected on the sensor. Therefore, resolving power numbers can be misleading in that they are assessed using a pure black and white high contrast line pattern. Therefore, any resolution number derived from such a method is not representative of resolution that can be observed in day-to-day scenes captured by the camera.
In addition, contrast can be affected by the photodiode themselves. Unlike pixels on your computer screen, which emit light from their whole surface area, sensor pixels are only sensitive to light on a portion of each pixel. The simplified sensor view in Figure 6 illustrates these areas called photodiode. Photodiodes are only sensitive to a certain range of brightness. Anything outside this range is either perceived as pure black or pure white. This range is called the dynamic range. A sensor’s dynamic range is greatly influenced by the photodiode size. The larger the photodiode, the higher the potential dynamic range. Essentially, a higher dynamic range camera will have more information, hence looses less contrast information in processing. As mentioned previously, usable resolution is affected by contrast drop. Hence, all else being equal, a camera equipped with larger sensor pixels should have a higher usable resolution.
Since the photodiode area is smaller than that of the pixel, the camera only has a fraction of the full image information to re-create the image in the picture area. This opens the door for the camera to misinterpret the photodiode information and introduce misinformation in the image. This will happen when the frequency of the line pattern exceeds the Nyquist frequency. This phenomenon is called aliasing and can be prevented with an anti-aliasing filter placed in front of the sensor.
Photodiodes are not sensitive to color – only brightness. To see colors, a filter, called a color filter array (CFA), is placed over each pixel of the sensor. The most common CFA uses the Bayer pattern illustrated in Figure 7.
In such a pattern, there are twice as many green filters as red or blue. When processing the raw information provided by each photodiode, the camera analyzes each pixel color information, and combined with that of adjacent pixels, recreates the full-color image. As this process introduced another variable, errors and resolution degradation can be introduced.
How to Make Sense of All This?
For those familiar with traditional film-based photography, my explanations may be comforting. The classical debate of 35mm versus medium format is very similar to the megapixels debate. As explained earlier in this article, more megapixels is not always better. The parallel to the classical debate will help us understand.
In traditional photography, the argument is as follows. It is possible for technically competent and well-equipped photographers to compete, in image quality, with a medium format image. How can this be?
For those not familiar with the debate, the basics are as follows. First, due to a larger image size, medium format lenses need a longer focal length for the same angle of coverage as 35mm lenses. Hence, medium format lenses of similar angle of coverage to 35mm typically are slower and have less depth of field due to the longer focal length. Second, medium format lenses are more difficult or pricier to design and manufacture with the same resolving power and contrast characteristics as a 35mm lens. To measure this resolution, traditional photographers typically use the IEEE standard STD 208-1995 “Measurement of Resolution of Camera Systems” target illustrated in Figure 8. The resolution figures are reported in line pairs per millimeter on the negative. This can mislead some photographers in that the number do not take into account the lower negative magnification required of the medium format.
To compensate for lens speed and shallow depth of field, medium format photographers tend to use smaller apertures and high speed film. The conclusion is that with slower film, lenses of a higher resolving power, tighter cropping, and good technique, it is possible for 35mm to compete with medium format. Bigger is not always better.
The digital camera scenario is similar in that megapixels are similar to film size. Hence, more megapixels do not always mean higher resolution.
The real measure of a digital camera is how many lines can a picture resolve. As mentioned earlier, resolution figures should be provided at both the center and corners of the images as well as for the three major directions, namely horizontal, vertical, and diagonal.
PIMA, now I3A, is currently developing an ISO standard to measure
digital camera resolution. This proposed International standard, ISO 12233
"Photography-Electronic still picture cameras - Resolution measurements"
uses the chart in Figure 9. The proposed measure of resolution is, correctly,
lines per image dimension. This method eliminates all variables such as the
number of megapixels by providing a common, meaningful, denominator: the image.
The method also alleviates any human subjectivity through the use of a plug-in
that automatically calculates resolution. As such it is similar to television
and video resolution specifications in lines per image height.
Reverting back to the resolution discussion, it can now be
safely assumed that statements referring to image quality and resolution in
printer’s dot per inch (dpi) is as erroneous and misleading as those using the
number of megapixels. Again, like megapixels, dpi should only be used as guidelines
and not absolute figures. Since pixels can be interpolated, it is the resolving
power of the camera (not image dimension) that will provide the image quality
· The number of megapixels is only an indicator of the potential image resolution – not absolute.
· If megapixel numbers must be used, picture pixels and output image should be used.
· To truly compare resolution, resolving power should be assessed in line pairs per image height (and width) using the ISO 12233 test chart and methodology under the same conditions. The figures should be derived for each variable (resolution direction: horizontal, vertical, and diagonal; location on the image: center and corner; focal length, aperture, scene brightness, ISO setting, and file format.)
· Evaluating resolution is not enough to guarantee image quality. Other very important factors such as dynamic range, noise, …, are also very important to ensure image quality.
· In some cases, users may want to trade resolution for features.
text is copyrighted 2002 and may not be copied, republished, redistributed,
or exploited in any manner without the express written permission of the author.
Test patterns illustrated in this article can be obtained at www.sinepatterns.com.
For a much more comprehensive exploration
of this subject
look at More — Understanding Resolution