EXTRACTOBJECTS Procedure

Cropping Detected Objects

In addition to enabling you to visualize object detections, PROC EXTRACTOBJECTS enables you to crop the detections and save them in an output table when you specify the METHOD=CROP option. An image can have multiple detections. The procedure provides two ways to save cropped detections in the output table by using the FORMAT= option in the OUTPUT statement:

FORMAT=MULTIPLECOLUMNS saves detections of an image in a single row and multiple columns. This means that there will be a single row with multiple columns that contain image binaries for each of the cropped objects. Each detected object is stored in a separate variable named _imageN_, where N identifies the Nth detected object. Each row in the output table corresponds to an image from the input image table. You can see a schema of the multiple-column format in Table 1. For an example that uses the multiple-column format, see Example 7.5.
FORMAT=MULTIPLEROWS stores detections of an image in multiple rows and a single column. This means that the image binaries of cropped objects are stored in separate rows in the output table. In the multiple-row format, the _parentId_ variable is added to the standard image table variables: _image_, _size_, _path_, _type_, _id_, and any other variables that you specify. You can see a schema that uses the multiple-row format in Table 2. For an example that uses the multiple-row format, see Example 7.4.

Table 1: Schema of the Multiple-Column Output

_path_	_id_	_image0_	_image1_	…	_imageN_
VARCHAR	Number	VARBINARY	VARBINARY	…	VARBINARY

Table 2: Schema of the Multiple-Row Output

_image_	_size_	_path_	_type_	_id_	_parentId_
VARBINARY	Number	VARCHAR	VARCHAR	Number	Number

Last updated: March 26, 2026