Face and Image analysis

Face descriptor

We proposed a new local image feature descriptor, Local Directional Pattern (LDP), which compute the edge response values in different directions and use these to encode the image texture.

  • An image descriptor which encode micro pattern.
  • Consistent in the presence of noise.
  • Robust for non monotonic illumination change.

 

Face recognition

The most critical aspect for any successful face recognition system is to find an efficient facial feature representation.

  • Able to encode texture micro patterns efficiently to detect a face.
  • As long as LDP shows consistency in random noise it shows better results.

 

Face expression

We proposed a code scheme that is robust against illumination, noise, and age conditions (in facial images). This code is Local Directional Pattern (LDP), and it is capable of recognize the expression of persons in previously mentioned conditions. The advantage of this method over previous work is that the edge response is more stable than illumination in variation conditions.

  • The edges responses are more stable than the intensity transitions.

 

Image enhancement

Sometimes images appear too dark or too bright to the viewer. Meaning the contrast of the image is not appropriate for human visual system. Changing this contrast can dramatically improve viewers’ comprehension of the image. But application that offers this kind of change ideally needs to be lossless, automatic, fast and reliable. We proposed three methods for image enhancement. One way is to consider the image block-wise. Traditional methods take the whole picture in consideration when performing enhancement. Weighted mixture of local and global transformation functions is another way to do contrast enhancement. We also proposed a method that allows dynamic histogram equalization. This Dynamic Histogram Equalization (DHE) technique takes control over the effect of traditional HE so that it performs the enhancement of n image without making any loss of details in it.

  • Works pretty well on the images with low dynamic range of gray levels.
  • No loss in image details.
  • Simple and computationally effective.

 

Image blockyness reduction

Video frames are always distorted by various artifacts at the time of acquisition or transmission. Error control is an important technique in image/video transmission over unreliable networks particularly in the wireless channel. In many video processing applications, accurate knowledge of the distortion level present in the input video sequence is very important for tuning the parameters of the corresponding video processing algorithm. With the rapid development of the application of video surveillance and broadcast systems, the evaluation of video quality becomes especially significant. The noise in a real system is mainly introduced by the camera and the quantization step as showed in Figure. But the distortions occur when the videos are transmitted through analog or digital medium. Some errors may be introduced when the analog video signal transmits in coaxial cable, but in wireless communication it cannot be ignored as occurring frequently.

  • Works pretty well for broadcast related distortions.
  • Light Weighted Human-perceivable measurement system.
  • Simple and computationally effective for broadcast and surveillance related applications.

 

Comments are closed.