22.15 A camera-feature wish list🔗⧉

A modern camera is a powerful computer wrapped around a very good sensor, and yet the list of things photographers keep asking it to do is strikingly long — and strikingly familiar, because most of the asks are the contents of this book. Lloyd Chambers, a reviewer with an unusually technical eye, publishes a recurring list of features camera makers ought to add; the author's blog appended a set of his own, chosen pointedly for being "easy and not that advanced computationally." That qualifier is the whole story. None of what follows needs a research breakthrough. It needs a manufacturer willing to expose the computer it already ships.

We group the wishes by theme. Each names the technique and points back to the chapter that develops it, so the appendix doubles as an index of the book's methods seen from the product side.

22.15.1 Exposure, ISO, and dynamic range🔗⧉

• Smarter auto-ISO [Durand]. Make the minimum shutter speed for auto-ISO a first-class, richly controllable setting — e.g. hold 1/200 s until ISO 3200, then accept 1/125 s up to a ceiling of 12800. This is pure policy logic over the exposure triangle (Image measurements as integrals); the camera has every number it needs and simply does not let you express the rule.
• Ambient/flash balance [Durand]. Give explicit control over how flash and ambient light are balanced — including the ability to switch between "match the ambient" and "keep a minimum shutter speed" behaviors, and to keep separate settings (a different max ISO, say) for flash versus no-flash shots. The computational cousin of this is flash/no-flash imaging, treated under computational illumination.
• In-camera HDR done right [Chambers]. Merge an exposure bracket into a single 16-bit raw in the camera, rather than a baked JPEG — the radiance-map merge of HDR merging performed before the file is written, leaving the photographer a high-dynamic-range raw to tone-map later (Tone mapping).
• Auto-ETTR [Chambers]. A true "expose to the right" mode that pushes the exposure as far up as the highlights allow, records by how much in metadata so the raw can be pulled back, handles specular highlights gracefully, and re-shoots if its estimate was wrong — metering as a small computation over the raw histogram (below), not a fixed matrix.
• Frame averaging for a clean long exposure [Durand/Chambers]. Capture N frames and average them in-camera to synthesize a low-noise, effectively low-ISO long exposure — the variance-reduction argument of Denoising (averaging N independent frames cuts the noise standard deviation by $\sqrt{N}$), offered as a shooting mode.

22.15.2 Bracketing more than exposure🔗⧉

• Alternate settings within a burst [Durand]. Let a burst cycle its settings shot to shot: flash / no-flash pairs (the input to flash/no-flash denoising); a slow-shutter frame for motion blur and subject tracking followed immediately by a fast "safe" frame; or a low-ISO/slow and high-ISO/fast pair to hedge the exposure. The camera fires the burst anyway — this asks only that consecutive frames differ by a chosen knob.
• Aperture bracketing for depth of field [Durand]. Bracket aperture while holding exposure constant, so the photographer can choose the depth of field after the fact — the controllable-DoF idea of Depth of field, and a poor man's step toward the post-capture refocus of a light-field camera.

22.15.3 Focus and depth of field🔗⧉

• Focus stacking, properly [Chambers]. Drive autofocus from near to far with control over the step size, accounting for field curvature, stopping at infinity, and detecting where the in-focus band lies — the capture half of the focus-stacking method in Depth of field (extend depth of field by combining many focal planes).
• Focus-shift compensation [Chambers]. Focus wide open (where focus is most precise) and compensate for the focus shift on stopping down — a small, well-understood optical correction the camera is positioned to make.
• Cycle through focus subjects [Durand]. When eye/face/subject autofocus locks onto the wrong person, give a one-press way to step through the other detected candidates — a UI affordance on top of the detection the autofocus system already runs (Focus, autofocus).

22.15.4 Computational raw and the sensor🔗⧉

• A true raw histogram [Chambers]. Show a histogram of the actual raw sensor data, per color channel, not one computed from the processed JPEG. Without it, ETTR and highlight judgment are guesswork; with it, they are arithmetic on the linear sensor signal (Basic image processing and ISP — the raw pipeline).
• Lossless-compressed raw [Chambers]. A genuinely lossless raw format — image coding applied to the mosaic before demosaicking.
• Pixel-shift high-resolution [Chambers]. A multi-shot mode that micro-shifts the sensor between frames and merges them for higher resolution and full-color (no demosaicking guess) per pixel, with intelligent handling of scene motion — the multi-frame super-resolution of Super-resolution and image priors built into the body.

22.15.5 Motion data, metadata, and workflow🔗⧉

• Record the IMU [Durand]. Log the gyroscope stream (for camera-shake removal, blur assessment, and video stabilization) and the accelerometer (for automatic perspective/leveling correction). The sensors are already in the camera for stabilization; saving their traces would feed IMU-aided deblurring and the stabilization pipeline of Video stabilization and rolling-shutter correction, and the keystone correction of Perspective distortion and its correction.
• Separate stills and video settings [Durand]. Keep independent exposure settings for stills and movies (1/200 s for stills, 1/30 s for video) so switching modes does not silently ruin one of them.
• In-camera rating that round-trips [Durand]. Let the photographer rate and sort during capture in a way that survives import into the editor (Lightroom et al.) — ratings written to standard metadata rather than a proprietary island (cross-ref Appendices#EXIF and image metadata, which covers XMP/IPTC, where ratings belong).
• Settings as a text file [Chambers]. Save and restore the full camera configuration to the card as a simple, human-readable file.

22.15.6 Panorama and multi-shot🔗⧉

• A real panorama mode [Chambers]. An electronic sweep-panorama mode that works like a phone's — capture, register, and stitch in-camera — i.e. the automatic stitching pipeline of Automatic panorama stitching from multiple views and feature matching, which standalone cameras have conspicuously ceded to phones.

22.15.7 The interface and the ecosystem🔗⧉

• Ergonomic basics [Chambers]. A self-timer with configurable duration and frame count that persists across power cycles; a one-press start/stop long exposure with an on-screen countdown; AF-ON decoupled from the shutter (back-button focus as a first-class mode).
• Get the pictures off the camera [Durand]. When the camera and the computer are on the same home Wi-Fi, they should simply figure out how to transfer — wireless offload that works without ceremony.
• Open the software ecosystem [Durand]. The wish that subsumes all the others: let third parties write apps for the camera. Almost every item above is a small program over data the camera already has. The reason your camera cannot do them is not that the math is hard — this book is largely the math, and it is not hard — but that the manufacturer has not written it and will not let anyone else. The bottleneck is openness, not optics. That is the lesson the whole list is really teaching, and a fitting note on which to close a book about everything a camera could compute.