Detection of prohibited objects concealed in an item, using image processing

1 . A system comprising one or more processing circuitries configured to:

obtain an image of an item acquired by an acquisition device,

perform a first detection using at least one first deep neural network, to detect at least one given area of the image comprising at least part of a given unprohibited concealing element of the item which enables concealment of a prohibited object, wherein the first detection is not directed to identifying prohibited objects or materials corresponding to prohibited objects,

determine that the given unprohibited concealing element belongs to a named category of unprohibited concealing elements associated with certain physical characteristic comprising at least one of material or shape, wherein a primary purpose of the given unprohibited concealing element comprises at least one of mechanical support, reinforcement, fastening, or identification,

use the named category of the given unprohibited concealing element to select a given second neural network among a plurality of different second neural networks operative to detect prohibited objects concealed using concealing elements,

wherein the given second neural network has been trained to detect in an image presence of a prohibited object concealed using a concealing element of said named category,

perform a second detection comprising using the given second neural network to detect whether the given area comprises a prohibited object, and

perform an action in response to detection of a presence of a prohibited object in the image.

2 . The system of claim 1 , wherein performing the second detection comprises detecting in the given area a prohibited object which is fully or at least partially concealed using the given unprohibited concealing element.

3 . The system of claim 1 , wherein performing the action includes triggering an alert.

4 . The system of claim 1 , configured to perform the second detection only on a fraction of the image including the given area.

5 . The system of claim 1 , configured to use the first deep neural network to determine the named category of the given unprohibited concealing element present in the given area.

6 . The system of claim 1 , wherein, for at least a subset of the plurality of different second neural networks, each given second neural network of the subset is trained to detect a prohibited object in an image comprising a different category of element enabling concealment of the prohibited object.

7 . The system of claim 6 , wherein each second neural network of the plurality of different second neural networks has been trained using a training set comprising:

images in which a prohibited object is concealed using an element of an item, the element being of a category for which the second neural network is trained, and

images in which no prohibited object is present.

8 . The system of claim 6 , wherein each second neural network of the plurality of different second neural networks has been trained using a training set comprising only images in which no prohibited object is present.

9 . The system of claim 1 , wherein the plurality of different second neural networks comprises:

a second neural network trained to detect in an image a prohibited object fully or at least partially concealed using an element of a first category;

another second neural network trained to detect in an image a prohibited object fully or at least partially concealed using an element of a second category, wherein the second category type is different from the first category.

10 . The system of claim 1 , wherein:

for at least a subset of the plurality of different second neural networks, each second neural network of the subset is trained to detect in an image presence of a prohibited object concealed using an element of a different category, and

the system is configured to select the given second neural network which is trained to detect in an image presence of a prohibited object concealed using an element of said named category.

11 . The system of claim 1 , configured to:

perform a detection in a majority of the image to detect whether a prohibited object is present in the image, thereby obtaining a first output informative of a presence of the prohibited object in the image,

perform the first detection and the second detection, thereby obtaining a second output informative of a presence of a prohibited object in the image,

perform an action in response to detection of a presence of a prohibited object in the image, wherein said detection is based at least on the first output and the second output.

12 . The system of claim 1 , wherein the given unprohibited concealing element comprises at least one of: baggage tube, baggage metallic tube, bag icon, combination lock, reinforcement element or metallic concealing element.

13 . A method comprising, by one or more processing circuitries:

obtaining an image of an item acquired by an acquisition device,

performing a first detection using at least one first deep neural network, to detect at least one given area of the image comprising at least part of a given unprohibited concealing element of the item which enables concealment of a prohibited object, wherein the first detection is not directed to identifying prohibited objects or materials corresponding to prohibited objects,

determining that the given unprohibited concealing element belongs to a named category of unprohibited concealing elements associated with certain physical characteristic comprising at least one of material or shape wherein a primary purpose of the given unprohibited concealing element comprises at least one of mechanical reinforcement, fastening or identification,

using the named category of the given unprohibited concealing element to select a given second neural network among a plurality of different second neural networks operative to detect prohibited objects concealed using concealing elements,

wherein the given second neural network has been trained to detect in an image presence of a prohibited object concealed using a concealing element of said named category,

performing a second detection comprising using the given second neural network to detect whether the given area comprises a prohibited object, and

performing an action in response to detection of a presence of a prohibited object in the image.

14 . The method of claim 13 , comprising performing at least one of (i) or (ii) or (iii) or (iv):

(i) performing the second detection comprises detecting in the given area a prohibited object which is fully or at least partially concealed using the given unprohibited concealing element, or

(ii) performing the second detection only on a fraction of the image including the given area, or

(iii) determining the named category of the given unprohibited concealing element is performed using the at least one first deep neural network, or

(iv) using the given second neural network to detect in the given area a prohibited object which is fully or at least partially concealed using the given unprohibited concealing element.

15 . The method of claim 13 , wherein at least one of (i) or (ii) is met:

(i) for at least a subset of the plurality of different second neural networks, each given second neural network of the subset is trained to detect a prohibited object in an image comprising a different category of element enabling concealment of the prohibited object, or

(ii) the plurality of different second neural networks comprises:

a second neural network trained to detect in an image a prohibited object fully or at least partially concealed using an element of a first category;

another second neural network trained to detect in an image a prohibited object fully or at least partially concealed using an element of a second category, wherein the second category is different from the first category.

16 . The method of claim 13 , wherein at least one of (i) or (ii) is met:

(i) each second neural network of the plurality of different second neural networks has been trained using a training set comprising:

images in which a prohibited object is concealed using an element of an item, the element being of a category for which the given second neural network is trained, and

images in which no prohibited object is present, or

(ii) each second neural network of the plurality of different second neural networks has been trained using a training set comprising only images in which no prohibited object is present.

17 . The method of claim 13 , comprising:

for at least a subset of a plurality of different second neural networks, each second neural network is trained to detect in an image presence of a prohibited object concealed using an element of a different category, wherein the method comprises selecting the given second neural network which is trained to detect in an image presence of a prohibited object concealed using an element of said named category.

18 . The method of claim 13 , comprising:

performing detection in a majority of the image to detect whether a prohibited object is present in the image, thereby obtaining a first output informative of a presence of the prohibited object in the image,

performing the first detection and the second detection, thereby obtaining a second output informative of a presence of a prohibited object in the image, and

performing an action in response to detection of a presence of a prohibited object in the image.

19 . The method of claim 13 , wherein the given unprohibited concealing element comprises at least one of: baggage tube, baggage metallic tube, bag icon, combination lock, reinforcement element or metallic concealing element.

20 . A non-transitory storage device readable by one or more processors, tangibly embodying a program of instructions executable by the one or more processors to perform:

obtaining an image of an item acquired by an acquisition device,

performing a first detection using at least one first deep neural network, to detect at least one given area of the image comprising at least part of a given unprohibited concealing element of the item which enables concealment of a prohibited object, wherein the first detection is not directed to identifying prohibited objects or materials corresponding to prohibited objects,

determining that the given unprohibited concealing element belongs to a named category of unprohibited concealing elements associated with certain physical characteristic comprising at least one of material or shape, wherein a primary purpose of the given unprohibited concealing element comprises at least one of mechanical support, reinforcement, fastening, or identification;

using the named category of the given unprohibited concealing element to select a given second neural network among a plurality of different second neural networks operative to detect prohibited objects concealed using concealing elements,

wherein the given second neural network has been trained to detect in an image presence of a prohibited object concealed using a concealing element of said named category,

performing a second detection comprising using the given second neural network to detect whether the given area comprises a prohibited object, and

performing an action in response to detection of a presence of a prohibited object in the image.