Image Compression for Target Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image processing apparatuses face reduced detection accuracy when dealing with high dynamic range image data converted to lower gradations, as the resolution in coarse luminance ranges is compromised, impairing the recognition of targets.
Innovation Solution
An image processing apparatus that acquires and compresses image data using nonlinear conversion characteristics, setting compression ratios to minimize resolution loss in recognition target ranges, ensuring accurate detection by maintaining contrast and resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If uniform compression is applied to convert from high dynamic range (12 bits) to standard dynamic range (8 bits), then the number of gradations is reduced to match processing performance, but the resolution in coarse luminance ranges is further lowered and detection accuracy is impaired
Solution Approach 1:
The patent applies different compression ratios to different luminance ranges. Specifically, it uses a first compression ratio for a first luminance range and a second compression ratio for a second luminance range, where the compression ratios are different. This local differentiation ensures that important luminance ranges maintain higher resolution while less critical ranges are compressed more aggressively, resolving the contradiction between overall compression efficiency and local detection accuracy.
Solution Approach 2:
The patent changes the compression ratio parameter based on the luminance range. By making the compression ratio a variable parameter that depends on the luminance range rather than a fixed uniform value, the system can optimize for both processing performance and detection accuracy in different regions of the image data.
2Device complexity
If gray scale conversion is performed to lower the number of gradations from 12 bits to 8 bits, then the image data can be processed by standard image processing apparatus, but the resolution is lowered and contrast in certain luminance ranges deteriorates
Solution Approach 1:
The patent applies different compression ratios to different luminance ranges. Specifically, it uses a first compression ratio for a first luminance range and a second compression ratio for a second luminance range, where the compression ratios are different. This local differentiation ensures that important luminance ranges maintain higher resolution while less critical ranges are compressed more aggressively, resolving the contradiction between overall compression efficiency and local detection accuracy.
Solution Approach 2:
The patent changes the compression ratio parameter based on the luminance range. By making the compression ratio a variable parameter that depends on the luminance range rather than a fixed uniform value, the system can optimize for both processing performance and detection accuracy in different regions of the image data.
Data Source
AI summary
In imaging processing, image data outputted from an imaging device is acquired as a data string formed of a first display value expressed by a first number of gradations obtained through conversion of a luminance value in accordance with preset nonlinear conversion characteristics. The first display value is compressed in accordance with preset compression characteristics, and outputted as a second display value expressed by a second number of gradations smaller than the first number of gradations. A recognition target is detected from an image expressed by compressed data that is a data string formed of the second display value. A ratio of the second number of gradations to the first number of gradations is taken as a basic compression ratio, a luminance range including at least a recognition target range that is a luminance range where the recognition target is estimated to be present is taken as a specified range, and the first display value corresponding to a boundary luminance value that is a minimum luminance value in the specified range is taken as a boundary first display value. In the recognition target range, the compression characteristics are set so that the second display value is a sum of a compressed value and the boundary first display value, the compressed value being obtained by compressing a value of not less than the boundary first display value among the first display values at a low compression ratio lower than the basic compression ratio.