Trim

Removes any borders or edges of an image which did does not change in color or transparency.

Crop down

Will crop the image at the right and the bottom edge to the next lower width and height which is divisible by eight.

This will lower the file size disproportionately higher in comparison to the pixels that get cropped.

Cover

Resizes the image to the next lower width and height which is divisible by eight.

To avoid a distorted image it is possible you will loose some pixel information at the sides (left/right or top/bottom).

Blurring radius (px)

The Selective Blur is used to blur a background so it can be better compressed.

Contrary to other blur algorithms, the Selective Blur doesn't act on all pixels: blur is applied only if the difference between its value and the value of the surrounding pixels is less than a defined threshold value. So, contrasts are preserved because difference is high on contrast limits.

Blurring threshold (%)

The Selective Blur is used to blur a background so it can be better compressed.

Contrary to other blur algorithms, the Selective Blur doesn't act on all pixels: blur is applied only if the difference between its value and the value of the surrounding pixels is less than a defined threshold value. So, contrasts are preserved because difference is high on contrast limits.

Keep metadata

Keeps the meta data (EXIF, IPTC, XMP) of the original image including ICC color profiles.

Color subsampling

Color subsampling is the practice of encoding images by implementing less resolution for color information than for brightness information, taking advantage of the human visual system‘s lower acuity for color differences than for brightness (from Wikipedia).

2x2 (4:2:0)
» smaller file size

Breaks the image into 2x2 pixel blocks and only stores the average color information for each 2x2 pixel group (default).

1x1 (4:4:4)
» bigger file size

The color information of every pixel will be stored.

More in my article Finally understanding JPG (#Chroma Subsampling)
and at https://en.wikipedia.org/wiki/Chroma_subsampling.

Color quality

Defines the quality for the color channels. Low quality results in wrong colors. Most of the time you can choose a value lower than the value of the brightness quality.

Amount of details (%)

Determines how many details are visible in the resulting image. Influences the quantization tables directly.

The graph shows all 64 values of a quantization table lined up in a row, following the zigzag pattern of the JPEG specification. On the X-axis, the first value therefore corresponds to the DC of the quantization table, followed by the AC values for the details of the image.

The gray line shows the values for a calculated quantization table (based on Appendix I of the JPEG specification) of a selected quality value, while the blue line shows the values of the actually used quantization table, which differs from the gray line if the "Amount of Details" slider was used. In this case, the quantization values for the detailed areas of the image are larger and thus the image is less detailed. To make it easier to see that larger values lead to a decrease in quality, the Y-axis has been reversed.

Practically this means: If you want to create the smallest possible image with good quality, first decrease the quality so far that you don't see any block artifacts anymore. Block artifacts are the most obvious and disturbing artifacts, so they should be avoided in the first place. Then reduce the amount of details as much as you find it acceptable.

Amount of details exponent

Influences how much the "Amount of Details" detail reduction is amplified for more detailed image regions (AC patterns).

Structure

Progressive

Image information will be stored from less details to more details. So a big image will be instantly visible but in low quality (default).

Baseline

Pixel information will be stored sequentially in the JPEG file. So a big image will get visible step by step in full resolution.

JPEG < 10 kB

Baseline (75% chance of smaller image)

JPEG > 10 kB

Progressive (94% chance of smaller image)

More in my article Finally understanding JPG (#Structure of the file).

Photoshop compatible

Allows to ensure the compatibility with Adobe Photoshop that cannot read the generated JPEGs. Results in a slighly larger file size.

Useful for photographers who need to open the JPEGs in Photoshop at a later time.

Use selective quality

Allows to have two different quality settings for different parts of the image.

Background color

Defines the background color if the transparency of the JPEG is not kept. Set the color in a hexadecimal format as you would do in e.g. in CSS.

Very useful if you have a transparent product image and want to flatten it on a defined background color.

Trellis Quantization

Trellis quantization is an adaptive quantization technique that select the set of levels in a transform block that minimises a rate-distortion metric.

Yes

Softer image, smaller file size (default).

No

Sharper image, bigger file size.

More at http://en.wikipedia.org/wiki/Trellis_quantization