Seamlines are used when mosaicking the raster data in a mosaic dataset. They are used instead of footprints to define the line along which the rasters in the mosaic dataset will be mosaicked when using the Seamline mosaic method.
When using the Seamline mosaic method, the mosaicked image will not change as you move around the image, as can happen when using mosaic methods like Closest To Center. Also, if you create seamlines, you can ensure the mosaicked image is the best possible mosaic by editing the seamlines so structures, such as buildings, are not along the line and that the seam between images is not obvious.
How the seamlines are created
To create seamlines for a mosaic dataset, you use the Build Seamlines tool. Seamlines are like footprints: there is one polygon representing each image. The shape of the polygon represents the part of the image that will be used to generate the mosaicked image when viewing the mosaic dataset. Once the seamlines are built, a Seamline layer will be present in the table of contents each time you add the mosaic dataset to ArcMap.
By default, the seamlines are generated using the North-West mosaic method. You can also choose to create the seamlines using the Closest To Viewpoint or By Attribute mosaic methods. Based on the mosaic method, a ranking value is stored in the SOrder field in the attribute table. When the Seamline mosaic method is used, the SOrder value is used instead of the ZOrder value. The SOrder value determines the order in which the images will be mosaicked together—like the ZOrder, the lower value is mosaicked on top. The illustration below displays the mosaic order of the three images and the area of overlap where the seamline will intersect.
There are four options when building seamlines. The first option is to create seamlines from the footprints, without computing a path between intersecting points, but considering the mosaic method (as shown below). This option is the GEOMETRY computation method. 
The second option is to build the seamlines by examining the values and patterns in the intersecting area and to compute a path between the intersecting points (above). This path is then merged with the footprint to create the seamline for each single image in the mosaic dataset (shown below). This option is the RADIOMETRY computation method.
The third option is to build seamlines by copying the footprints where each seamline is just the copy of the corresponding footprint. This option is the COPY_FOOTPRINT computation method.
The fourth option copies an existing seamline of a raster item to its siblings which share the same group name. This option is the COPY_TO_SIBLING computation method. This is commonly used with satellite imagery when the panchromatic band does not always share the same extent as the multispectral band. This option makes sure they share the same seamline.
The diagram below
represents a series of seamlines created for 12 images. The mosaicking order uses the North-West sorting method and the RADIOMETRY computation method.
Using your own seamlines
If you have your own seamline (polygon) file, you need to first run the Build Seamline tool to establish the framework for the seamlines (using the COPY_FOOTPRINT computation method). You can then use the Import Mosaic Dataset Geometry tool to replace the seamlines with your own.
Generating seamlines at different dataset levels
If your mosaic dataset is comprised of different resolutions of data, such as high-resolution orthophotography and low-resolution satellite imagery, you can generate seamlines that are unique to each of these datasets by specifying an appropriate cell size. When you don't specify a cell size the seamlines are automatically generated for each cell size specified in the cell size level table of the mosaic dataset.
Blending along seamlines
You can define a blending value (feathering) and type that will occur along the seamline. Blending occurs between pixels where there are overlapping rasters. By default, the Blend Width is defined in the mosaic dataset's default properties. This value defines the distance in pixels (at the display scale). You can override this value in the seamline attribute table.
The seamline attribute table contains a Blend Width and Blend Type field for every seamline in the mosaic dataset. The options for the Blend Type are Both, Inside, and Outside. If the Blend Width value is 10, and you use Both as the blend type, then 5 pixels will be blended on the inside and outside of the seamline. If the value is 10, and the blend type is Inside, then 10 pixels will be blended on the inside of the seamline. You can override the Blend Type value in the seamline attribute table. If you are using Field Calculator to edit the values, specify values of 1 for Both, 2 for Inside, and 3 for Outside.
In order to blend the pixel values along a seamline the mosaic method must be Seamline and the mosaic operator must be set to Blend. This can be set in the default properties of the mosaic dataset, or chosen by the user.
Removing seamlines
To remove all the seamlines from a mosaic dataset, right-click the mosaic dataset in ArcCatalog or the Catalog window and click Remove > Remove Seamlines.