Image data pre-processing for neural networks

1. An image processor comprising:

an input module for receiving high dynamic range image data from an image sensor, the image data comprising a plurality of pixel data, each having characteristic data associated therewith;

an image processing module arranged to perform at least one image processing operation on at least some of the pixel data of the high dynamic range image data to generate processed image data;

a characteristic processing module arranged to perform at least one characteristic processing operation on at least some of the characteristic data to generate processed characteristic data, wherein the at least some of the characteristic data is associated with the at least some of the pixel data of the high dynamic range image data used to generate the processed image data;

an exposure processing module arranged to generate an exposure mask, the exposure mask being based on at least one exposure boundary of the high dynamic range image data; and

an output module for outputting the processed image data, the exposure mask, and the processed characteristic data to a neural processing unit arranged to generate an output using at least one trained neural network;

wherein:

the at least one characteristic processing operation performed on the at least some of the characteristic data corresponds to the at least one image processing operation performed on the at least some of the pixel data of the high dynamic range image data;

the exposure processing module is arranged to generate the exposure mask in parallel when performing the image processing operation and the characteristic processing operation; and

the processed image data, the exposure mask, and processed characteristic data being configured for use by the at least one trained neural network.

2. The image processor of claim 1 , wherein the portion of the image data is a pixel of the image data.

3. The image processor of claim 1 , wherein the portion of the image data is a predefined region of the image data.

4. The image processor of claim 1 , wherein the at least one image processing operation adjusts the image data, and the at least one characteristic processing operations adjust the characteristic data; wherein the adjustments to the image data and the characteristic data are substantially the same.

5. A method comprising the steps of:

obtaining high dynamic range image data from an image sensor, the high dynamic range image data comprising a plurality of pixel data, each having characteristic data associated therewith;

applying at least one image processing operation to at least some of the pixel data of the image data to generate processed image data;

applying at least one characteristic processing operation to at least some of the characteristic data to generate processed characteristic data, wherein the at least some of the characteristic data is associated with the at least some the pixel data of the high dynamic range image data used to generate the processed image data;

determining exposure data, wherein the exposure data is associated with at least one exposure boundary of the high dynamic range image data;

generating an exposure mask based on the exposure data; and

outputting the processed image data, the exposure mask, and the processed characteristic data for subsequent processing by a neural processing unit, the neural processing unit arranged to generate an output using at least one trained neural network;

wherein:

the at least one characteristic processing operation to be performed on the portion at least some of the characteristic data corresponds to the at least one image processing operation performed on the at least some of the pixel data of the high dynamic range image data;

the generation of the exposure mask, the at least one image processing operation, and the at least one characteristic processing operation are performed parallel; and

the processed image data, the exposure mask, and processed characteristic data being configured for use by the at least one trained neural network.

6. The method of claim 5 , wherein the portion of the image data is a pixel of the image data.

7. The method of claim 5 , wherein the portion of the image data is a predefined region of the image data.

8. The method of claim 5 , wherein the at least one image processing operation adjusts the image data, and the at least one characteristic processing operations adjust the characteristic data, wherein the adjustments to the characteristic data and the image data are substantially the same.

9. A system comprising:

an image sensor for capturing high dynamic range image data, the high dynamic range image data comprising a plurality of pixel data, each having characteristic data associated therewith;

an image signal processor for:

receiving the image data from the image sensor;

applying at least one image processing operation to at least some of the pixel data of the high dynamic range image data to produce processed image data;

applying at least one characteristic processing operation to at least a portion some of the characteristic data, to produce processed characteristic data, wherein:

the at least some of the characteristic data is associated with the at least some of the pixel data of the high dynamic range image data used to produce the processed image data; and

wherein the at least one characteristic processing operation applied to the at least some of the characteristic data corresponds to the at least one image processing operation applied to the at least some of the pixel data of the high dynamic range image data;

determining exposure data, wherein the exposure data is associated with at least one exposure boundary of the high dynamic range image data;

generating an exposure mask based on the exposure data, wherein the generation of the exposure mask, the at least one image processing operation, and the at least one characteristic processing operation are performed parallel; and

a neural processing unit for generating an output using at least one trained neural network, wherein the at least one trained neural network receives the processed image data, the exposure mask, and the processed characteristic data as an input, the processed image data and processed characteristic data being configured for use by the at least one trained neural network.