26
SSIP 2003 Project suggestions

SSIP 2003 Project suggestions. Project 1 Label the group photo- locate and identify faces and label them. * Input group photo ( for example 10 people)

Embed Size (px)

Citation preview

SSIP 2003

Project suggestions

Project 1 Label the group photo- locate and identify faces and

label them.   * Input group photo ( for example 10 people) * Segment it to isolate people/faces * Number the faces * Extract the faces * Build of library of faces * From photos of similar faces try to find that person

on the group photo

Remarks: Difficulty medium

Project 1 Label the group photo- locate and identify faces and

label them.   * Input group photo ( for example 10 people) * Segment it to isolate people/faces * Number the faces * Extract the faces * Build of library of faces * From photos of similar faces try to find that person

on the group photo

Remarks: Difficulty medium

Project 2 Door entry system- feature analysis of a face using point separation /

wavelets   Input: images of several faces  Operation: Identify key points (end of nose/ ears/ mouth). Measure

distances and angles between these (for different orientations). Feed results into a statistical analysis routine. Identify for unknown image most likely match. Alternatively use the wavelet transform to generate ‘spectrum’ and identify key ‘frequencies’. Then do statistical matching. Wavelet transform (see Numerical recipes) needs to be downloaded). An alternative would be images of hands.

  Output: Demo of door entry type system based on photo.  Coding: as desired (but not development of GUI)  Remarks: Difficulty quite hard.

Project 3 Block world line drawing generated from range image input of

scene.  Input: 2-D Image of 3-d block world as a range image where

pixel value is a function of distance (or simulation) from digital/ video camera. Suggestion, generate synthetic data first, then try out on real data.

  Operation: Identification of lines and corners, linking of this to

extract model of scene.  Output: Line drawing of scene. Rotation of object to create

new scene.  Remarks: Difficulty medium.

Range image

12

10

22

Project 3 Block world line drawing generated from range image input of

scene.  Input: 2-D Image of 3-d block world as a range image where

pixel value is a function of distance (or simulation) from digital/ video camera. Suggestion, generate synthetic data first, then try out on real data.

  Operation: Identification of lines and corners, linking of this to

extract model of scene.  Output: Line drawing of scene. Rotation of object to create

new scene.  Remarks: Difficulty medium.

Project 6 Barley seeds classification Input: a set of photos Task: Design a classifier and test its performances Output: classes and statistical results Difficulty: medium

Extract relevant shape and texture features from barley seeds images. Design a classifier for the feature vectors obtained and test its performances. Use the "live one out" testing method, as the number of samples in the experiment is very low.

Project 6 Barley seeds classification Input: a set of photos Task: Design a classifier and test its performances Output: classes and statistical results Difficulty: medium

Extract relevant shape and texture features from barley seeds images. Design a classifier for the feature vectors obtained and test its performances. Use the "live one out" testing method, as the number of samples in the experiment is very low.

Project 7 Extraction of 'skeleton‘, or the medial axis transform,

from 3-D objects.  Input: Generate simulated 3-d objects  Method: Add noise, Extract 'skeleton' that is medial

axis. Determine characteristics of medial axis to identify type of object (tube. cylinder, torous etc).

  Output: 'skeleton' as a 3D object. Classification of

objects.  Remarks: Difficulty medium to hard

Project 8Raster to vector conversionStarting from a reasonably complex

drawing saved in raster format, decompose it in the composite vectors from which it was formed

Input- raster imageOutput vector descriptionDifficult- fairly easy

Project 9 Registration of 2-D images using critical point matching.  Input: Aerial photos and maps, [or CT and MR slices].  Operation: extraction of contour, identification of (geometrical)

critical points (maximum gradient, 2nd differential etc of contour), computation of transformation to register two such images (e.g. by SVD). Extension to non-rigid body matching would be possible.

  Output: Transformed image (and subtraction image)  Coding: as desired (but not GUI)  Remarks: Difficulty medium

Project 10 Match of fragment of coastline to map  Starting off with a segment of coastline (or river with bridges) from

a map, of different scale and noise properties extracted from a (much) larger segment, perform a best fit to identify the section of coastline. One method that could be used would be by correlation of a chain code representation. Creation of chain code or equivalent from map segment is part of the project. Imagine satellite images

  Input: 2d map  Output: contour with match identified  Coding: as desired  Difficulty: quite easy

Project 11 Motion tracking and correction  Given a set of images in time which are similar but not

identical, derive a method for identifying the motion that has occurred (in 2d) between different images (shift and rotation) such that they can be adjusted and corrected. An example would be an image of a footballer (or a cell) on a sequence of images.

  Input: Starting from image sequence, identify features for

example with skeleton, or determine regions of change by subtraction etc. Find significant changes. [This is the same as finding moving object is constant background scene.]

  Output: image marked with regions of change

Project 12 2-D edge detection using cost minimization/ snakes.  Input: Medical and/or other images   Operation: Compute gradient image. Define a transform, for example

polar, a cost function, for example circumference and gradient. Minimize path in transformed data by cost minimization. Alternative, use a snake for example using greedy algorithm. The object is to find an algorithm to link the points identified on a gradient map to give continuous enclosing contours. Think out extension to 3d [important].

  Output: Image with contour. Algorithm to identify organ, for example

left ventricle of heart, without manual intervention.  Coding: In C/C++ (or under PIP) in form which could be used in

package  Remarks: Difficulty medium. Problem is robustness

Project 13 Segmentation of a fuzzy region by clustering and pixel affinity.  Input: a suitable 2-D data set, for example a satellite photo of

mountains in Afghanistan, or an MR brain slice.  Operation: given a seed point generated manually, and a criteria

defining affinity, for example a voxel values, gradient values and distance, grow the region to extract the selected object (cave entrances in mountains/ MS lesions). The connection should be on the basis on likelihood that the neighboring voxels can be linked to the target voxel (e.g. by building up a linked list/ tree). Manual editing could be added. For MR the problem can be extended into 3D.

  Output: color coded labeled map.  Remarks: Difficulty reasonably easy.

Project 14 Segmentation using texture  Input: satellite images of regions with different

crops  Method: Detection of different regions using a

texture parameter, attempting to assess what type of texture exists (grass, forest etc)

  Output: Labeled image  Difficulty: Easy to medium

Project 15 Traffic sign identification  Input: Images taken from a car (take care about

motion, lights, shadows, perspective views); a set of traffic signs

  Method: Select features used for recognition, find a

candidate sign on the image; identify it .  Output: The identifier of the found trafic sign, (and

the next few similar ones).   Difficulty: Easy to medium

Project 16 Queuing theory demonstration  Input: None Method: Demonstrate graphically illustration of

queuing theory. A good example would be a simulation of road traffic flow, to illustrate wave phenomena (standing and moving waves) associated with partial obstructions.

Output: Graphical demo, preferably in form of 2d image/ map [along lines of Sim city with graphs].

Difficulty: variable

Project 17 Determination of an ellipsoid or other geometric shape

in 3D from 3 projections only  Input: Simulated projections  Operation: fitting 2D data to 3D constrained object at

various angles, adding noise, convolving with system point function to include blurring. Mathematic model required.

  Output: Parameters of shape  Coding: As desired but modular  Remarks: Difficulty fairly easy.

Project 18 AUTOMATIC POKER PLAYER  Input: Images of cards, number of example images for

initial development have been provided  Locate the playing cards in a scene. Classify them.

Possibly look into classifying cards partly obscured by other cards

  Output: Parameters of shape  Coding: As desired but modular  Remarks: Difficulty fairly easy.

Select 3 or 4 projects you would like to work on

Give the list of these projects (with your name) until Friday 12:00 to Prof. Cucu

The teams will be formed on the base of these lists and announced on Friday at 15:00

The teams present their work at the end of SSIP (on Friday, 11 July)

The first 3 teams will get big prizes and certificates (and they will be mentioned –as usual – on the home page of SSIP)

Good luck!